πΉ Video: https://youtu.be/M8tqVF4Gyi0
\nπ Presentation can be downloaded from here: https://t.me/sberlogabio/59283
Welcome to the webinar on Friday: John Mitchell who is expert both in bioinformatics and machine learning, and also experienced Kaggler and one of the top participants of CAFA5 , will share his experience on the past Kaggle competition \"Novoenzymes\" which in certain respects similar to CAFA5 challenge.
\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"
\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)
The zoom link will be posted here 5 minutes before the start.
\n\nAbstract:
\nKaggle's Novozymes Enzyme Stability Prediction was a challenging competition that rewarded expertise in bioinformatics as much as in Machine Learning. Two issues that made the competition particularly difficult were that the training data was rather different from the test data, and that there was no obvious or easy local validation protocol available.
\nA wide variety of features and models proved valuable, while ensembling was generally considered essential. The nature of the competition lent itself to overfitting, and it was unsurprising that there was a large shake-up between Public LB and Private LB ranks.
\nI discuss the experience gained from this competition and consider how the lessons learned might be applicable to other bioinformatic competitions.
Everybody is welcome !
\nVideo record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\n=============================================
\nπΉ Video: https://youtu.be/4W12qYMuMhk
\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"
\nβοΈ Thursday 27 July, 19.00 (CET time)
The talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).
\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.
The zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga
\n==============================================
\nπΉ Video: https://youtu.be/eGk2Y3zUe3Q
\nEverybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.
\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"
\nβοΈ Monday 17 July, 17.00 (CET time)
Sequence-structure based bioinformatics approach is a new direction of bioinformatics which is based on representation of sequences as their structural formula and analysis of structure-property relationships using molecular descriptors and machine learning algorithms. Some cases using of this approach will be discussed during webinar.
\nAbout the reporter: Prof. A. Lagunin (https://scholar.google.ru/citations?user=qaSuEUkAAAAJ&hl ) authored numerous publications in the field of bioinformatics.
\nThe zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\n=====================================================================
\nπΉ Video: https://youtu.be/ccyILqjlXZY
\n π¨βπ¬Trifonova Sofya \"HMMER application to CAFA5.\"
\nβοΈ Friday 14 July, 17.00 (CET time)
HMMER ( http://hmmer.org/ ) is a sensitive method for finding protein domains (https://en.wikipedia.org/wiki/Protein_domain) . Domains in proteins define it functional features, that is why using this method could help in better predictive ability of ML models in CAFA5.
\nFor example, one may try to create 0-1 features enumerated by proteins domains: protein have particular domain (1) or not (0). And try to add them as predicators to the task.
\nSofya created and shared several notebooks that one in particular:
\nhttps://www.kaggle.com/code/sonyagrey/domains-encoding?scriptVersionId=136451490&cellId=9 (please upvote).
The zoom link will be posted here 10-15 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\nPS
\n\"Pham\" is database with protein domains: http://pfam.xfam.org/ (new link: https://www.ebi.ac.uk/interpro/ )
\nSee also Kaggle dataset with Pham from Google AI: https://www.kaggle.com/datasets/googleai/pfam-seed-random-split
\n========================================================
\nPrevious webinar:
\nYou are welcome to a webinar on Friday, though it is NOT directly related to CAFA5 challenge, however one might think how to apply these ideas to the challenge. May be it can be used for data augmentation - if the predictions of mutation effects are precise enough one can try to augment data by mutated proteins and their changed functions - though seems such ideas not so easy to implement and outcomes are so clear at the moment. The webinar is for research/educational purposes.
\nπ¨βπ¬ Marina Pak \"ProDDG (https://ivankovlab.ru/proddg)- database of data on protein stability change upon mutation (ΞΞG) and predictors of mutation effect.\"
\nβοΈ Friday 7 July, 19.00 (CET time)
\nπΉ Video record: https://youtu.be/uCABm3ymFDg
It was an interesting research webinar, though we were no able find ways how to directly apply these ideas to CAFA5 challenge.
\n\nProDDG - https://ivankovlab.ru/proddg - is a web-service for developers, assessors and users of tools for predicting the effect of protein mutations. Using ProDDG you can:
\nI will talk about features of ProDDG and what problems the service solves, demonstrate use cases and share my experience of development of such projects.
\nThe zoom link will be posted here 10-15 minutes before the start. Video record will be on youtube later.
\n========================================================
\nPrevious webinar:
\nπΉ Video record of the webinar: https://youtu.be/hpYfLLqyo9Y
\nπ Presentation: https://docs.google.com/presentation/d/1nxpFDVvxGb7F9uJ92swvNkXIUzzOm4Vi1qY1J0AoTUU/edit?usp=drivesdk
\n(see also attachment to that post)
\nπ¨βπ¬ Liza Geraseva \"A short review of methods used in CAFA challenges\"
\nβοΈ Monday 3 July, 18.00 (CET time, e.g. Paris time)
The zoom link will be posted here 10-15 minutes before the start. Video record will be on youtube later.
\nAbstract:
\nThe Critical Assessment of protein Function Annotation algorithms (CAFA) is an experiment designed to provide a large-scale assessment of computational methods dedicated to predicting protein function, using a time challenge. CAFA1-3 challenges have released a large number of approaches that can be be improved and used in CAFA5. I am going to review some of these approaches, concentrating on those developed by Hoehndorf lab as most successful.
PS
\nLiza Geraseva shared many remarkable notebooks for the competition (please upvote):
\nhttps://www.kaggle.com/code/geraseva/keops-knn
\nhttps://www.kaggle.com/code/geraseva/protbert-embedding
\nhttps://www.kaggle.com/code/geraseva/deepgo
\nAnd several others see: https://www.kaggle.com/geraseva/code
========================================================
\nπΉ Video record of the webinar: https://youtu.be/h4zTjPzPud4
\nπ Presentation: https://docs.google.com/presentation/d/1VrYMvJPlSrPd56pNXYm5PUylnMLEp9d0oMjcobOP47o/edit#slide=id.p
========================================================
\nThe competition is important for the research community. Thanks for the organizers and Kaggle !
\nTo combine these two themes, my colleagues and me plan to organize some educational/research activity around that competition. The ideas are: to facilitate discussions, networking, knowledge sharing widely around that research topic and ideally to help to convert Kaggler's findings to some research papers, as well as to give some help to those who are less experienced but willing to participate.
\nLet us emphasize - it is not intended to be private sharing / or even sharing of high scorable approaches, but rather sharing knowledge WIDELY around that research topic / that competition / competitions and data science in general. Whatever done will be publicly shared here on Kaggle forum.
Everybody is welcome to participate !
\nTo start: we plan an introductory webinar mainly for those who have not yet started. By Andrey Shevtsov ( who created and shared ESM2 embeddings: https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/406168 - please upvote ! )
\nπ¨βπ¬ Andrey Shevtsov \"Introduction to CAFA5 Protein function prediction Kaggle competition\"
\nβοΈ Thursday 22 June, 19.00 (CET time, e.g. Paris time)
\nπΉ Video record of the webinar: https://youtu.be/h4zTjPzPud4
\nπ Presentation: https://docs.google.com/presentation/d/1VrYMvJPlSrPd56pNXYm5PUylnMLEp9d0oMjcobOP47o/edit#slide=id.p
The zoom link will be posted here 10-15 minutes before the start.
\nAbstract:
\nThe competition to predict the functions of CAFA5 proteins continues at Kaggle (https://www.kaggle.com/competitions/cafa-5-protein-function-prediction). There are still two months to take part in it. Welcome to participate in our seminar at 22.06.2023 to listen to the introductory information on the competition.
\nProteins are chains of amino acids (letters) folded into 3d structure. They perform many important functions in cells. The protein acquires its functions already in the folded state (in some cases - after additional modifications). But there is an opinion that the spatial arrangement of a protein molecule depends entirely on the original chain/sequence. This opinion is indirectly confirmed by the fact that there are methods for predicting the 3d structure of a protein by its sequence (like Alphafold/ESM family of models), as well as various encoders for performing downstream tasks. In this competition, the task is to predict protein properties by its sequence. The properties are presented in the form of directed acyclic graphs GO (gene ontology (http://geneontology.org/docs/ontology-documentation)). Interestingly, bioinformatic approaches are visible in the discussion.
πΉ Video: https://youtu.be/M8tqVF4Gyi0
\nπ Presentation can be downloaded from here: https://t.me/sberlogabio/59283
Welcome to the webinar on Friday: John Mitchell who is expert both in bioinformatics and machine learning, and also experienced Kaggler and one of the top participants of CAFA5 , will share his experience on the past Kaggle competition \"Novoenzymes\" which in certain respects similar to CAFA5 challenge.
\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"
\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)
The zoom link will be posted here 5 minutes before the start.
\n\nAbstract:
\nKaggle's Novozymes Enzyme Stability Prediction was a challenging competition that rewarded expertise in bioinformatics as much as in Machine Learning. Two issues that made the competition particularly difficult were that the training data was rather different from the test data, and that there was no obvious or easy local validation protocol available.
\nA wide variety of features and models proved valuable, while ensembling was generally considered essential. The nature of the competition lent itself to overfitting, and it was unsurprising that there was a large shake-up between Public LB and Private LB ranks.
\nI discuss the experience gained from this competition and consider how the lessons learned might be applicable to other bioinformatic competitions.
Everybody is welcome !
\nVideo record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\n=============================================
\nπΉ Video: https://youtu.be/4W12qYMuMhk
\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"
\nβοΈ Thursday 27 July, 19.00 (CET time)
The talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).
\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.
The zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga
\n==============================================
\nπΉ Video: https://youtu.be/eGk2Y3zUe3Q
\nEverybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.
\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"
\nβοΈ Monday 17 July, 17.00 (CET time)
Sequence-structure based bioinformatics approach is a new direction of bioinformatics which is based on representation of sequences as their structural formula and analysis of structure-property relationships using molecular descriptors and machine learning algorithms. Some cases using of this approach will be discussed during webinar.
\nAbout the reporter: Prof. A. Lagunin (https://scholar.google.ru/citations?user=qaSuEUkAAAAJ&hl ) authored numerous publications in the field of bioinformatics.
\nThe zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\n=====================================================================
\nπΉ Video: https://youtu.be/ccyILqjlXZY
\n π¨βπ¬Trifonova Sofya \"HMMER application to CAFA5.\"
\nβοΈ Friday 14 July, 17.00 (CET time)
HMMER ( http://hmmer.org/ ) is a sensitive method for finding protein domains (https://en.wikipedia.org/wiki/Protein_domain) . Domains in proteins define it functional features, that is why using this method could help in better predictive ability of ML models in CAFA5.
\nFor example, one may try to create 0-1 features enumerated by proteins domains: protein have particular domain (1) or not (0). And try to add them as predicators to the task.
\nSofya created and shared several notebooks that one in particular:
\nhttps://www.kaggle.com/code/sonyagrey/domains-encoding?scriptVersionId=136451490&cellId=9 (please upvote).
The zoom link will be posted here 10-15 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\nPS
\n\"Pham\" is database with protein domains: http://pfam.xfam.org/ (new link: https://www.ebi.ac.uk/interpro/ )
\nSee also Kaggle dataset with Pham from Google AI: https://www.kaggle.com/datasets/googleai/pfam-seed-random-split
\n========================================================
\nPrevious webinar:
\nYou are welcome to a webinar on Friday, though it is NOT directly related to CAFA5 challenge, however one might think how to apply these ideas to the challenge. May be it can be used for data augmentation - if the predictions of mutation effects are precise enough one can try to augment data by mutated proteins and their changed functions - though seems such ideas not so easy to implement and outcomes are so clear at the moment. The webinar is for research/educational purposes.
\nπ¨βπ¬ Marina Pak \"ProDDG (https://ivankovlab.ru/proddg)- database of data on protein stability change upon mutation (ΞΞG) and predictors of mutation effect.\"
\nβοΈ Friday 7 July, 19.00 (CET time)
\nπΉ Video record: https://youtu.be/uCABm3ymFDg
It was an interesting research webinar, though we were no able find ways how to directly apply these ideas to CAFA5 challenge.
\n\nProDDG - https://ivankovlab.ru/proddg - is a web-service for developers, assessors and users of tools for predicting the effect of protein mutations. Using ProDDG you can:
\nI will talk about features of ProDDG and what problems the service solves, demonstrate use cases and share my experience of development of such projects.
\nThe zoom link will be posted here 10-15 minutes before the start. Video record will be on youtube later.
\n========================================================
\nPrevious webinar:
\nπΉ Video record of the webinar: https://youtu.be/hpYfLLqyo9Y
\nπ Presentation: https://docs.google.com/presentation/d/1nxpFDVvxGb7F9uJ92swvNkXIUzzOm4Vi1qY1J0AoTUU/edit?usp=drivesdk
\n(see also attachment to that post)
\nπ¨βπ¬ Liza Geraseva \"A short review of methods used in CAFA challenges\"
\nβοΈ Monday 3 July, 18.00 (CET time, e.g. Paris time)
The zoom link will be posted here 10-15 minutes before the start. Video record will be on youtube later.
\nAbstract:
\nThe Critical Assessment of protein Function Annotation algorithms (CAFA) is an experiment designed to provide a large-scale assessment of computational methods dedicated to predicting protein function, using a time challenge. CAFA1-3 challenges have released a large number of approaches that can be be improved and used in CAFA5. I am going to review some of these approaches, concentrating on those developed by Hoehndorf lab as most successful.
PS
\nLiza Geraseva shared many remarkable notebooks for the competition (please upvote):
\nhttps://www.kaggle.com/code/geraseva/keops-knn
\nhttps://www.kaggle.com/code/geraseva/protbert-embedding
\nhttps://www.kaggle.com/code/geraseva/deepgo
\nAnd several others see: https://www.kaggle.com/geraseva/code
========================================================
\nπΉ Video record of the webinar: https://youtu.be/h4zTjPzPud4
\nπ Presentation: https://docs.google.com/presentation/d/1VrYMvJPlSrPd56pNXYm5PUylnMLEp9d0oMjcobOP47o/edit#slide=id.p
========================================================
\nThe competition is important for the research community. Thanks for the organizers and Kaggle !
\nTo combine these two themes, my colleagues and me plan to organize some educational/research activity around that competition. The ideas are: to facilitate discussions, networking, knowledge sharing widely around that research topic and ideally to help to convert Kaggler's findings to some research papers, as well as to give some help to those who are less experienced but willing to participate.
\nLet us emphasize - it is not intended to be private sharing / or even sharing of high scorable approaches, but rather sharing knowledge WIDELY around that research topic / that competition / competitions and data science in general. Whatever done will be publicly shared here on Kaggle forum.
Everybody is welcome to participate !
\nTo start: we plan an introductory webinar mainly for those who have not yet started. By Andrey Shevtsov ( who created and shared ESM2 embeddings: https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/406168 - please upvote ! )
\nπ¨βπ¬ Andrey Shevtsov \"Introduction to CAFA5 Protein function prediction Kaggle competition\"
\nβοΈ Thursday 22 June, 19.00 (CET time, e.g. Paris time)
\nπΉ Video record of the webinar: https://youtu.be/h4zTjPzPud4
\nπ Presentation: https://docs.google.com/presentation/d/1VrYMvJPlSrPd56pNXYm5PUylnMLEp9d0oMjcobOP47o/edit#slide=id.p
The zoom link will be posted here 10-15 minutes before the start.
\nAbstract:
\nThe competition to predict the functions of CAFA5 proteins continues at Kaggle (https://www.kaggle.com/competitions/cafa-5-protein-function-prediction). There are still two months to take part in it. Welcome to participate in our seminar at 22.06.2023 to listen to the introductory information on the competition.
\nProteins are chains of amino acids (letters) folded into 3d structure. They perform many important functions in cells. The protein acquires its functions already in the folded state (in some cases - after additional modifications). But there is an opinion that the spatial arrangement of a protein molecule depends entirely on the original chain/sequence. This opinion is indirectly confirmed by the fact that there are methods for predicting the 3d structure of a protein by its sequence (like Alphafold/ESM family of models), as well as various encoders for performing downstream tasks. In this competition, the task is to predict protein properties by its sequence. The properties are presented in the form of directed acyclic graphs GO (gene ontology (http://geneontology.org/docs/ontology-documentation)). Interestingly, bioinformatic approaches are visible in the discussion.
π¨βπ¬ Liza Geraseva \"A short review of methods used in CAFA challenges\"
\nβοΈ Monday 3 July, 18.00 (CET time, e.g. Paris time)
The zoom link will be posted here 10-15 minutes before the start. Video record will be on youtube later.
\nAbstract:
\nThe Critical Assessment of protein Function Annotation algorithms (CAFA) is an experiment designed to provide a large-scale assessment of computational methods dedicated to predicting protein function, using a time challenge. CAFA1-3 challenges have released a large number of approaches that can be be improved and used in CAFA5. I am going to review some of these approaches, concentrating on those developed by Hoehndorf lab as most successful.
PS
\nLiza Geraseva shared many remarkable notebooks for the competition (please upvote):
\nhttps://www.kaggle.com/code/geraseva/keops-knn
\nhttps://www.kaggle.com/code/geraseva/protbert-embedding
\nhttps://www.kaggle.com/code/geraseva/deepgo
\nAnd several others see: https://www.kaggle.com/geraseva/code
Zoom link: https://us02web.zoom.us/j/84239724288?pwd=L3hQSXRTaUppWlpmWEs5SUFzMitUZz09
", "rawMarkdown": "\nZoom link: https://us02web.zoom.us/j/84239724288?pwd=L3hQSXRTaUppWlpmWEs5SUFzMitUZz09\n\n", "votes": 2 }, { "id": 2328476, "postDate": "2023-07-03T15:52:59.613Z", "content": "Thank for zoom link:)
\nhttps://us02web.zoom.us/j/84239724288?pwd=L3hQSXRTaUppWlpmWEs5SUFzMitUZz09
", "rawMarkdown": "Thank for zoom link:)\n\nhttps://us02web.zoom.us/j/84239724288?pwd=L3hQSXRTaUppWlpmWEs5SUFzMitUZz09", "votes": 2 } ] }, { "id": 2392430, "postDate": "2023-08-15T17:17:20.520Z", "content": "πΉ Video record: https://youtu.be/M8tqVF4Gyi0
\nπ Presentation: https://t.me/sberlogabio/59283
Welcome to the webinar on Friday: John Mitchell who is expert both in bioinformatics and machine learning, and also experienced Kaggler and one of the top participants of CAFA5 , will share his experience on the past Kaggle competition \"Novoenzymes\" which in certain respects similar to CAFA5 challenge.
\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"
\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)
The zoom link will be posted here 5 minutes before the start.
\n\nAbstract:
\nKaggle's Novozymes Enzyme Stability Prediction was a challenging competition that rewarded expertise in bioinformatics as much as in Machine Learning. Two issues that made the competition particularly difficult were that the training data was rather different from the test data, and that there was no obvious or easy local validation protocol available.
\nA wide variety of features and models proved valuable, while ensembling was generally considered essential. The nature of the competition lent itself to overfitting, and it was unsurprising that there was a large shake-up between Public LB and Private LB ranks.
\nI discuss the experience gained from this competition and consider how the lessons learned might be applicable to other bioinformatic competitions.
Everybody is welcome !
\nVideo record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
", "rawMarkdown": "πΉ Video record: https://youtu.be/M8tqVF4Gyi0\nπ Presentation: https://t.me/sberlogabio/59283\n\nWelcome to the webinar on Friday: [John Mitchell ](https://www.kaggle.com/jbomitchell) who is expert both in bioinformatics and machine learning, and also experienced Kaggler and one of the top participants of CAFA5 , will share his experience on the past Kaggle competition \"Novoenzymes\" which in certain respects similar to CAFA5 challenge.\n\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)\n\nThe **zoom link** will be posted here 5 minutes before the start.\n\n[Add to google calendar\n](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230818T160000Z%2F20230818T173000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=%20John%20Mitchell%20%20%22Kaggle%20Competition%20Review%3A%20Novozymes%20Enzyme%20Stability%20Prediction%22)\n\n**Abstract:**\nKaggle's Novozymes Enzyme Stability Prediction was a challenging competition that rewarded expertise in bioinformatics as much as in Machine Learning. Two issues that made the competition particularly difficult were that the training data was rather different from the test data, and that there was no obvious or easy local validation protocol available.\nA wide variety of features and models proved valuable, while ensembling was generally considered essential. The nature of the competition lent itself to overfitting, and it was unsurprising that there was a large shake-up between Public LB and Private LB ranks.\nI discuss the experience gained from this competition and consider how the lessons learned might be applicable to other bioinformatic competitions.\n\nEverybody is welcome !\n\nVideo record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga", "votes": 3 }, { "id": 2341826, "postDate": "2023-07-12T10:11:24.040Z", "content": "π¨βπ¬ Trifonova Sofya \"HMMER application to CAFA5.\"
\nβοΈ Friday 14 July, 17.00 (CET time)
HMMER ( http://hmmer.org/ ) is a sensitive method for finding protein domains (https://en.wikipedia.org/wiki/Protein_domain) . Domains in proteins define it functional features, that is why using this method could help in better predictive ability of ML models in CAFA5.
\nFor example, one may try to create 0-1 features enumerated by proteins domains: protein have particular domain (1) or not (0). And try to add them as predicators to the task.
\nSofya created and shared several notebooks that one in particular:
\nhttps://www.kaggle.com/code/sonyagrey/domains-encoding?scriptVersionId=136451490&cellId=9 (please upvote).
The zoom link will be posted here 10-15 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
", "rawMarkdown": "π¨βπ¬ Trifonova Sofya \"HMMER application to CAFA5.\"\nβοΈ Friday 14 July, 17.00 (CET time) \n\n[Add to google calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230714T140000Z%2F20230714T163000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=Trifonova%20Sofya%20%20%22HMMER%20application%20to%20CAFA5.%22)\n\nHMMER ( http://hmmer.org/ ) is a sensitive method for finding protein domains (https://en.wikipedia.org/wiki/Protein_domain) . Domains in proteins define it functional features, that is why using this method could help in better predictive ability of ML models in CAFA5.\n\nFor example, one may try to create 0-1 features enumerated by proteins domains: protein have particular domain (1) or not (0). And try to add them as predicators to the task. \n\nSofya created and shared several notebooks that one in particular: \nhttps://www.kaggle.com/code/sonyagrey/domains-encoding?scriptVersionId=136451490&cellId=9 (please upvote). \n\nThe **zoom link** will be posted here 10-15 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga \n", "votes": 4 }, { "id": 2311366, "postDate": "2023-06-21T06:13:24.687Z", "content": "@serangu @simonveitner @zmcxjt Thank you for your kind words ! Would you be so kind to contact me al.chervov@gmail.com
", "rawMarkdown": "@serangu @simonveitner @zmcxjt Thank you for your kind words ! Would you be so kind to contact me al.chervov@gmail.com ", "votes": 3 }, { "id": 2310179, "postDate": "2023-06-20T07:05:42.967Z", "content": "I am happy to share any insights I have gained so far in the field of automatic protein function prediction. I am also planning on presenting my preliminary work on using protein language models to predict function at ISMB/ECCB next month.
", "rawMarkdown": "I am happy to share any insights I have gained so far in the field of automatic protein function prediction. I am also planning on presenting my preliminary work on using protein language models to predict function at ISMB/ECCB next month.", "votes": 3 }, { "id": 2310111, "postDate": "2023-06-20T05:52:26.620Z", "content": "That's a great idea! Hope I can make it.
", "rawMarkdown": "That's a great idea! Hope I can make it.", "votes": 3 }, { "id": 2310088, "postDate": "2023-06-20T05:24:19.557Z", "content": "@alexandervc good idea! This is great for the community. The topic of the competition is difficult for Data Scientists. And the introductory lecture will help many!
", "rawMarkdown": "@alexandervc good idea! This is great for the community. The topic of the competition is difficult for Data Scientists. And the introductory lecture will help many!", "votes": 3 }, { "id": 2313626, "postDate": "2023-06-22T19:29:35.380Z", "content": "i liked the meeting much today and i hope we can repeat it soon.
", "rawMarkdown": "i liked the meeting much today and i hope we can repeat it soon.", "votes": 4, "replies": [ { "id": 2314181, "postDate": "2023-06-23T08:42:06.147Z", "content": "Thanks Simon ! And thanks a lot for sharing your notebooks and discussion !
\nWould be so kind to contact me al.chervov@gmail.com - if you have an opportunity we will schedule your talk - brief or long - up to you
Presentation is attached
\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"
\nβοΈ Thursday 27 July
πΉ Video: https://youtu.be/4W12qYMuMhk
", "rawMarkdown": "Presentation is attached\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"\nβοΈ Thursday 27 July \n\nπΉ Video: https://youtu.be/4W12qYMuMhk", "votes": 2 }, { "id": 2361875, "postDate": "2023-07-27T16:52:59.457Z", "content": "Zoom link for today webinar:
\nhttps://us02web.zoom.us/j/85437399900?pwd=QXZjcGNHaXpiYjRtUTlNMzZ1WXZDdz09
π¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"
\nβοΈ Thursday 27 July, 19.00 (CET time)
Add to Google Calendar
\nThe talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).
\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.
The zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga
", "rawMarkdown": "Zoom link for today webinar:\nhttps://us02web.zoom.us/j/85437399900?pwd=QXZjcGNHaXpiYjRtUTlNMzZ1WXZDdz09\n\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"\nβοΈ Thursday 27 July, 19.00 (CET time)\n\nAdd to Google Calendar\n\nThe talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.\n\nThe zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga\n\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga\n\n", "votes": 2 }, { "id": 2361238, "postDate": "2023-07-27T09:27:40.610Z", "content": "Reminder - today webinar:
\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"
\nβοΈ Thursday 27 July, 19.00 (CET time)
Add to Google Calendar
\nThe talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).
\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.
The zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga
", "rawMarkdown": "Reminder - today webinar:\n\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"\nβοΈ Thursday 27 July, 19.00 (CET time)\n\nAdd to Google Calendar\n\nThe talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.\n\nThe zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga\n\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga\n\n", "votes": 2 }, { "id": 2357419, "postDate": "2023-07-24T19:12:39.017Z", "content": "π¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"
\nβοΈ Thursday 27 July, 19.00 (CET time)
The talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).
\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.
The zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga
", "rawMarkdown": "π¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"\nβοΈ Thursday 27 July, 19.00 (CET time)\n\n[Add to Google Calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230727T170000Z%2F20230727T193000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=Henrik%20Nielsen%2CVineet%20Thumuluri%20%2C%20Jos%C3%A9%20Juan%20Almagro%20Armenteros%20%20%22DeepLoc%202.0%3A%20multi-label%20subcellular%20localization%20prediction%20using%20protein%20language%20models%22)\n\nThe talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.\n\nThe zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga\n\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga", "votes": 2 }, { "id": 2349343, "postDate": "2023-07-18T10:21:03.423Z", "content": "πΉ Video: https://youtu.be/eGk2Y3zUe3Q
\nEverybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.
\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"
\nβοΈ Monday 17 July, 17.00 (CET time)
Sequence-structure based bioinformatics approach is a new direction of bioinformatics which is based on representation of sequences as their structural formula and analysis of structure-property relationships using molecular descriptors and machine learning algorithms. Some cases using of this approach will be discussed during webinar.
\nAbout the reporter: Prof. A. Lagunin (https://scholar.google.ru/citations?user=qaSuEUkAAAAJ&hl ) authored numerous publications in the field of bioinformatics.
", "rawMarkdown": "πΉ Video: https://youtu.be/eGk2Y3zUe3Q \n\nEverybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.\n\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"\nβοΈ Monday 17 July, 17.00 (CET time) \n\n\n Sequence-structure based bioinformatics approach is a new direction of bioinformatics which is based on representation of sequences as their structural formula and analysis of structure-property relationships using molecular descriptors and machine learning algorithms. Some cases using of this approach will be discussed during webinar. \n\nAbout the reporter: Prof. A. Lagunin (https://scholar.google.ru/citations?user=qaSuEUkAAAAJ&hl ) authored numerous publications in the field of bioinformatics.", "votes": 2 }, { "id": 2348350, "postDate": "2023-07-17T14:55:45.907Z", "content": "https://us02web.zoom.us/j/89020453099?pwd=aXRGUkRwUXBSaHozSk5QU1lsV0NPQT09
\nEverybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.
\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"
\nβοΈ Monday 17 July, 17.00 (CET time)
Everybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.
\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"
\nβοΈ Monday 17 July, 17.00 (CET time)
Sequence-structure based bioinformatics approach is a new direction of bioinformatics which is based on representation of sequences as their structural formula and analysis of structure-property relationships using molecular descriptors and machine learning algorithms. Some cases using of this approach will be discussed during webinar.
\nAbout the reporter: Prof. A. Lagunin (https://scholar.google.ru/citations?user=qaSuEUkAAAAJ&hl ) authored numerous publications in the field of bioinformatics.
\nThe zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
", "rawMarkdown": "Everybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.\n\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"\nβοΈ Monday 17 July, 17.00 (CET time)\n\n[Add to Google Calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230717T150000Z%2F20230717T173000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=Prof.%20A.Lagunin%20%22Sequence-structure%20based%20bioinformatics.%22)\n\nSequence-structure based bioinformatics approach is a new direction of bioinformatics which is based on representation of sequences as their structural formula and analysis of structure-property relationships using molecular descriptors and machine learning algorithms. Some cases using of this approach will be discussed during webinar.\n\nAbout the reporter: Prof. A. Lagunin (https://scholar.google.ru/citations?user=qaSuEUkAAAAJ&hl ) authored numerous publications in the field of bioinformatics.\n\nThe zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga", "votes": 2 }, { "id": 2344545, "postDate": "2023-07-14T14:56:07.800Z", "content": "https://us02web.zoom.us/j/81847029643?pwd=bkkvYUhESmViemFTdExRMDdyMDYzQT09
\nπ¨βπ¬ Trifonova Sofya \"HMMER application to CAFA5.\"
\nβοΈ Friday 14 July, 17.00 (CET time)
\n\nπ¨βπ¬ Trifonova Sofya \"HMMER application to CAFA5.\"
\n
\n βοΈ Friday 14 July, 17.00 (CET time)
Looks like the Google calendar invitation for July 14 is for 16:00 CET? Anyhow, I'm looking forward to it.
\nIddo
", "rawMarkdown": "> π¨βπ¬ Trifonova Sofya \"HMMER application to CAFA5.\"\n> βοΈ Friday 14 July, 17.00 (CET time)\n\n\nLooks like the Google calendar invitation for July 14 is for 16:00 CET? Anyhow, I'm looking forward to it.\n\nIddo", "votes": 2, "replies": [ { "id": 2344445, "postDate": "2023-07-14T14:03:16.897Z", "content": "Sorry if so (((
\nIt is planned on 17.00 CET.
Zoom link:
\nhttps://us02web.zoom.us/j/88968811599?pwd=YitNc0RaeFVRSmNqemxoQ2dKWnJZZz09
\nπ¨βπ¬ Marina Pak \"ProDDG (https://ivankovlab.ru/proddg)- database of data on protein stability change upon mutation (ΞΞG) and predictors of mutation effect.\"
\nβοΈ Friday 7 July, 19.00 (CET time)
Thank you for recording
\nπΉ Video: https://youtu.be/hpYfLLqyo9Y
\nβοΈ Abstract: https://t.me/sberlogabig/278
πΉ Video: https://youtu.be/M8tqVF4Gyi0
\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"
\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)
Zoom link:
\nhttps://us02web.zoom.us/j/84399146917?pwd=TitKVm5NbS9tWTNrcm96SjY1dnFhZz09
Welcome to the webinar on Friday: John Mitchell who is expert both in bioinformatics and machine learning, and also experienced Kaggler and one of the top participants of CAFA5 , will share his experience on the past Kaggle competition \"Novoenzymes\" which in certain respects similar to CAFA5 challenge.
\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"
\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)
Zoom meetings will start in half an hour?
", "rawMarkdown": "Zoom meetings will start in half an hour?" }, { "id": 2313413, "postDate": "2023-06-22T16:50:42.270Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 2325292, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2023-07-01T08:29:21.993000", "content": "π¨βπ¬ Liza Geraseva \"A short review of methods used in CAFA challenges\"
\nβοΈ Monday 3 July, 18.00 (CET time, e.g. Paris time)
The zoom link will be posted here 10-15 minutes before the start. Video record will be on youtube later.
\nAbstract:
\nThe Critical Assessment of protein Function Annotation algorithms (CAFA) is an experiment designed to provide a large-scale assessment of computational methods dedicated to predicting protein function, using a time challenge. CAFA1-3 challenges have released a large number of approaches that can be be improved and used in CAFA5. I am going to review some of these approaches, concentrating on those developed by Hoehndorf lab as most successful.
PS
\nLiza Geraseva shared many remarkable notebooks for the competition (please upvote):
\nhttps://www.kaggle.com/code/geraseva/keops-knn
\nhttps://www.kaggle.com/code/geraseva/protbert-embedding
\nhttps://www.kaggle.com/code/geraseva/deepgo
\nAnd several others see: https://www.kaggle.com/geraseva/code
Zoom link: https://us02web.zoom.us/j/84239724288?pwd=L3hQSXRTaUppWlpmWEs5SUFzMitUZz09
", "votes": 2, "replies": [] }, { "id": 2328476, "author_name": "", "author_url": "", "post_date": "2023-07-03T15:52:59.613000", "content": "Thank for zoom link:)
\nhttps://us02web.zoom.us/j/84239724288?pwd=L3hQSXRTaUppWlpmWEs5SUFzMitUZz09
", "votes": 2, "replies": [] } ] }, { "id": 2392430, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2023-08-15T17:17:20.520000", "content": "πΉ Video record: https://youtu.be/M8tqVF4Gyi0
\nπ Presentation: https://t.me/sberlogabio/59283
Welcome to the webinar on Friday: John Mitchell who is expert both in bioinformatics and machine learning, and also experienced Kaggler and one of the top participants of CAFA5 , will share his experience on the past Kaggle competition \"Novoenzymes\" which in certain respects similar to CAFA5 challenge.
\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"
\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)
The zoom link will be posted here 5 minutes before the start.
\n\nAbstract:
\nKaggle's Novozymes Enzyme Stability Prediction was a challenging competition that rewarded expertise in bioinformatics as much as in Machine Learning. Two issues that made the competition particularly difficult were that the training data was rather different from the test data, and that there was no obvious or easy local validation protocol available.
\nA wide variety of features and models proved valuable, while ensembling was generally considered essential. The nature of the competition lent itself to overfitting, and it was unsurprising that there was a large shake-up between Public LB and Private LB ranks.
\nI discuss the experience gained from this competition and consider how the lessons learned might be applicable to other bioinformatic competitions.
Everybody is welcome !
\nVideo record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
", "votes": 3, "replies": [] }, { "id": 2341826, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2023-07-12T10:11:24.040000", "content": "π¨βπ¬ Trifonova Sofya \"HMMER application to CAFA5.\"
\nβοΈ Friday 14 July, 17.00 (CET time)
HMMER ( http://hmmer.org/ ) is a sensitive method for finding protein domains (https://en.wikipedia.org/wiki/Protein_domain) . Domains in proteins define it functional features, that is why using this method could help in better predictive ability of ML models in CAFA5.
\nFor example, one may try to create 0-1 features enumerated by proteins domains: protein have particular domain (1) or not (0). And try to add them as predicators to the task.
\nSofya created and shared several notebooks that one in particular:
\nhttps://www.kaggle.com/code/sonyagrey/domains-encoding?scriptVersionId=136451490&cellId=9 (please upvote).
The zoom link will be posted here 10-15 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
", "votes": 4, "replies": [] }, { "id": 2311366, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2023-06-21T06:13:24.687000", "content": "@serangu @simonveitner @zmcxjt Thank you for your kind words ! Would you be so kind to contact me al.chervov@gmail.com
", "votes": 3, "replies": [] }, { "id": 2310179, "author_name": "ZMC", "author_url": "", "post_date": "2023-06-20T07:05:42.967000", "content": "I am happy to share any insights I have gained so far in the field of automatic protein function prediction. I am also planning on presenting my preliminary work on using protein language models to predict function at ISMB/ECCB next month.
", "votes": 3, "replies": [] }, { "id": 2310111, "author_name": "Simon Veitner", "author_url": "", "post_date": "2023-06-20T05:52:26.620000", "content": "That's a great idea! Hope I can make it.
", "votes": 3, "replies": [] }, { "id": 2310088, "author_name": "serangu", "author_url": "", "post_date": "2023-06-20T05:24:19.557000", "content": "@alexandervc good idea! This is great for the community. The topic of the competition is difficult for Data Scientists. And the introductory lecture will help many!
", "votes": 3, "replies": [] }, { "id": 2313626, "author_name": "Simon Veitner", "author_url": "", "post_date": "2023-06-22T19:29:35.380000", "content": "i liked the meeting much today and i hope we can repeat it soon.
", "votes": 4, "replies": [ { "id": 2314181, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2023-06-23T08:42:06.147000", "content": "Thanks Simon ! And thanks a lot for sharing your notebooks and discussion !
\nWould be so kind to contact me al.chervov@gmail.com - if you have an opportunity we will schedule your talk - brief or long - up to you
Presentation is attached
\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"
\nβοΈ Thursday 27 July
πΉ Video: https://youtu.be/4W12qYMuMhk
", "votes": 2, "replies": [] }, { "id": 2361875, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2023-07-27T16:52:59.457000", "content": "Zoom link for today webinar:
\nhttps://us02web.zoom.us/j/85437399900?pwd=QXZjcGNHaXpiYjRtUTlNMzZ1WXZDdz09
π¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"
\nβοΈ Thursday 27 July, 19.00 (CET time)
Add to Google Calendar
\nThe talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).
\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.
The zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga
", "votes": 2, "replies": [] }, { "id": 2361238, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2023-07-27T09:27:40.610000", "content": "Reminder - today webinar:
\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"
\nβοΈ Thursday 27 July, 19.00 (CET time)
Add to Google Calendar
\nThe talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).
\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.
The zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga
", "votes": 2, "replies": [] }, { "id": 2357419, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2023-07-24T19:12:39.017000", "content": "π¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"
\nβοΈ Thursday 27 July, 19.00 (CET time)
The talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).
\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.
The zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga
", "votes": 2, "replies": [] }, { "id": 2349343, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2023-07-18T10:21:03.423000", "content": "πΉ Video: https://youtu.be/eGk2Y3zUe3Q
\nEverybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.
\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"
\nβοΈ Monday 17 July, 17.00 (CET time)
Sequence-structure based bioinformatics approach is a new direction of bioinformatics which is based on representation of sequences as their structural formula and analysis of structure-property relationships using molecular descriptors and machine learning algorithms. Some cases using of this approach will be discussed during webinar.
\nAbout the reporter: Prof. A. Lagunin (https://scholar.google.ru/citations?user=qaSuEUkAAAAJ&hl ) authored numerous publications in the field of bioinformatics.
", "votes": 2, "replies": [] }, { "id": 2348350, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2023-07-17T14:55:45.907000", "content": "https://us02web.zoom.us/j/89020453099?pwd=aXRGUkRwUXBSaHozSk5QU1lsV0NPQT09
\nEverybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.
\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"
\nβοΈ Monday 17 July, 17.00 (CET time)
Everybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.
\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"
\nβοΈ Monday 17 July, 17.00 (CET time)
Sequence-structure based bioinformatics approach is a new direction of bioinformatics which is based on representation of sequences as their structural formula and analysis of structure-property relationships using molecular descriptors and machine learning algorithms. Some cases using of this approach will be discussed during webinar.
\nAbout the reporter: Prof. A. Lagunin (https://scholar.google.ru/citations?user=qaSuEUkAAAAJ&hl ) authored numerous publications in the field of bioinformatics.
\nThe zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga
", "votes": 2, "replies": [] }, { "id": 2344545, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2023-07-14T14:56:07.800000", "content": "https://us02web.zoom.us/j/81847029643?pwd=bkkvYUhESmViemFTdExRMDdyMDYzQT09
\nπ¨βπ¬ Trifonova Sofya \"HMMER application to CAFA5.\"
\nβοΈ Friday 14 July, 17.00 (CET time)
\n\nπ¨βπ¬ Trifonova Sofya \"HMMER application to CAFA5.\"
\n
\n βοΈ Friday 14 July, 17.00 (CET time)
Looks like the Google calendar invitation for July 14 is for 16:00 CET? Anyhow, I'm looking forward to it.
\nIddo
", "votes": 2, "replies": [ { "id": 2344445, "author_name": "Alexander Chervov", "author_url": "", "post_date": "2023-07-14T14:03:16.897000", "content": "Sorry if so (((
\nIt is planned on 17.00 CET.
Zoom link:
\nhttps://us02web.zoom.us/j/88968811599?pwd=YitNc0RaeFVRSmNqemxoQ2dKWnJZZz09
\nπ¨βπ¬ Marina Pak \"ProDDG (https://ivankovlab.ru/proddg)- database of data on protein stability change upon mutation (ΞΞG) and predictors of mutation effect.\"
\nβοΈ Friday 7 July, 19.00 (CET time)
Thank you for recording
\nπΉ Video: https://youtu.be/hpYfLLqyo9Y
\nβοΈ Abstract: https://t.me/sberlogabig/278
πΉ Video: https://youtu.be/M8tqVF4Gyi0
\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"
\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)
Zoom link:
\nhttps://us02web.zoom.us/j/84399146917?pwd=TitKVm5NbS9tWTNrcm96SjY1dnFhZz09
Welcome to the webinar on Friday: John Mitchell who is expert both in bioinformatics and machine learning, and also experienced Kaggler and one of the top participants of CAFA5 , will share his experience on the past Kaggle competition \"Novoenzymes\" which in certain respects similar to CAFA5 challenge.
\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"
\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)
Zoom meetings will start in half an hour?
", "votes": 0, "replies": [] }, { "id": 2313413, "author_name": "", "author_url": "", "post_date": "2023-06-22T16:50:42.270000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2309759": "πΉ Video: https://youtu.be/M8tqVF4Gyi0\nπ Presentation can be downloaded from here: https://t.me/sberlogabio/59283\n\nWelcome to the webinar on Friday: [John Mitchell ](https://www.kaggle.com/jbomitchell)who is expert both in bioinformatics and machine learning, and also experienced Kaggler and one of the top participants of CAFA5 , will share his experience on the past Kaggle competition [\"Novoenzymes\" ](https://www.kaggle.com/competitions/novozymes-enzyme-stability-prediction/overview) which in certain respects similar to CAFA5 challenge. \n\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)\n\nThe **zoom link** will be posted here 5 minutes before the start. \n\n[Add to google calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230818T160000Z%2F20230818T173000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=%20John%20Mitchell%20%20%22Kaggle%20Competition%20Review%3A%20Novozymes%20Enzyme%20Stability%20Prediction%22)\n\n**Abstract:**\nKaggle's Novozymes Enzyme Stability Prediction was a challenging competition that rewarded expertise in bioinformatics as much as in Machine Learning. Two issues that made the competition particularly difficult were that the training data was rather different from the test data, and that there was no obvious or easy local validation protocol available. \nA wide variety of features and models proved valuable, while ensembling was generally considered essential. The nature of the competition lent itself to overfitting, and it was unsurprising that there was a large shake-up between Public LB and Private LB ranks. \nI discuss the experience gained from this competition and consider how the lessons learned might be applicable to other bioinformatic competitions. \n\nEverybody is welcome ! \n\nVideo record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga \n\n\n=============================================\n\nπΉ Video: https://youtu.be/4W12qYMuMhk\n\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"\nβοΈ Thursday 27 July, 19.00 (CET time)\n\n[Add to Google Calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230727T170000Z%2F20230727T193000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=Henrik%20Nielsen%2CVineet%20Thumuluri%20%2C%20Jos%C3%A9%20Juan%20Almagro%20Armenteros%20%20%22DeepLoc%202.0%3A%20multi-label%20subcellular%20localization%20prediction%20using%20protein%20language%20models%22)\n\nThe talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ). \nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.\n\nThe **zoom link** will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga \n\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga\n\n==============================================\n\nπΉ Video: https://youtu.be/eGk2Y3zUe3Q \n\nEverybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.\n\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"\nβοΈ Monday 17 July, 17.00 (CET time) \n\n\n[Add to Google Calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230717T150000Z%2F20230717T173000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=Prof.%20A.Lagunin%20%22Sequence-structure%20based%20bioinformatics.%22)\n\n Sequence-structure based bioinformatics approach is a new direction of bioinformatics which is based on representation of sequences as their structural formula and analysis of structure-property relationships using molecular descriptors and machine learning algorithms. Some cases using of this approach will be discussed during webinar. \n\nAbout the reporter: Prof. A. Lagunin (https://scholar.google.ru/citations?user=qaSuEUkAAAAJ&hl ) authored numerous publications in the field of bioinformatics.\n\nThe **zoom link** will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga \n\n=====================================================================\n\n\nπΉ Video: https://youtu.be/ccyILqjlXZY\n π¨βπ¬Trifonova Sofya \"HMMER application to CAFA5.\"\nβοΈ Friday 14 July, 17.00 (CET time) \n\nHMMER ( http://hmmer.org/ ) is a sensitive method for finding protein domains (https://en.wikipedia.org/wiki/Protein_domain) . Domains in proteins define it functional features, that is why using this method could help in better predictive ability of ML models in CAFA5.\n\nFor example, one may try to create 0-1 features enumerated by proteins domains: protein have particular domain (1) or not (0). And try to add them as predicators to the task. \n\nSofya created and shared several notebooks that one in particular: \nhttps://www.kaggle.com/code/sonyagrey/domains-encoding?scriptVersionId=136451490&cellId=9 (please upvote). \n\nThe **zoom link** will be posted here 10-15 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga \n\nPS \n\n\"Pham\" is database with protein domains: http://pfam.xfam.org/ (new link: https://www.ebi.ac.uk/interpro/ )\n\nSee also Kaggle dataset with Pham from Google AI: https://www.kaggle.com/datasets/googleai/pfam-seed-random-split\n\n\n\n========================================================\n\nPrevious webinar: \n\nYou are welcome to a webinar on Friday, though it is NOT directly related to CAFA5 challenge, however one might think how to apply these ideas to the challenge. May be it can be used for data augmentation - if the predictions of mutation effects are precise enough one can try to augment data by mutated proteins and their changed functions - though seems such ideas not so easy to implement and outcomes are so clear at the moment. The webinar is for research/educational purposes. \n\nπ¨βπ¬ Marina Pak \"ProDDG (https://ivankovlab.ru/proddg)- database of data on protein stability change upon mutation (ΞΞG) and predictors of mutation effect.\"\nβοΈ Friday 7 July, 19.00 (CET time) \nπΉ Video record: https://youtu.be/uCABm3ymFDg\n\nIt was an interesting research webinar, though we were no able find ways how to directly apply these ideas to CAFA5 challenge.\n\n\n[Add to google calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230707T170000Z%2F20230707T183000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=Marina%20Pak%20%28SkolTech%29%20%20%22ProDDG%20%20%28https%3A%2F%2Fivankovlab.ru%2Fproddg%29-%20%D0%B1%D0%B0%D0%B7%D0%B0%20%D0%B4%D0%B0%D0%BD%D0%BD%D1%8B%D1%85%20%D1%8D%D1%84%D1%84%D0%B5%D0%BA%D1%82%D0%BE%D0%B2%20%D0%BC%D1%83%D1%82%D0%B0%D1%86%D0%B8%D0%B9%20%D0%BD%D0%B0%20%D1%81%D1%82%D0%B0%D0%B1%D0%B8%D0%BB%D1%8C%D0%BD%D0%BE%D1%81%D1%82%D1%8C%20%D0%B1%D0%B5%D0%BB%D0%BA%D0%B0%20%28%CE%94%CE%94G%29%20%D0%B8%20%D0%BF%D1%80%D0%B5%D0%B4%D1%81%D0%BA%D0%B0%D0%B7%D0%B0%D1%82%D0%B5%D0%BB%D0%B5%D0%B9%20%CE%94%CE%94G.%22)\n\nProDDG - https://ivankovlab.ru/proddg - is a web-service for developers, assessors and users of tools for predicting the effect of protein mutations. Using ProDDG you can:\n\n- search, filter, analyze and download ddG data\n- download ready-to-use ddG datasets for training, testing and assessment\n- find overlaps between ΞΞG datasets based on protein sequence identities and compile leakage-free datasets\n- discover popular tools for prediction of mutation effect\n\nI will talk about features of ProDDG and what problems the service solves, demonstrate use cases and share my experience of development of such projects.\n\nThe **zoom link** will be posted here 10-15 minutes before the start. Video record will be on youtube later.\n\n========================================================\n\nPrevious webinar: \n\nπΉ Video record of the webinar: https://youtu.be/hpYfLLqyo9Y\nπ Presentation: https://docs.google.com/presentation/d/1nxpFDVvxGb7F9uJ92swvNkXIUzzOm4Vi1qY1J0AoTUU/edit?usp=drivesdk\n(see also attachment to that post)\nπ¨βπ¬ Liza Geraseva \"A short review of methods used in CAFA challenges\"\nβοΈ Monday 3 July, 18.00 (CET time, e.g. Paris time) \n\n[Add to google calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230703T160000Z%2F20230703T183000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=Liza%20Geraseva%20%22A%20short%20review%20of%20methods%20used%20in%20CAFA%20challenges%22)\n\nThe **zoom link** will be posted here 10-15 minutes before the start. Video record will be on youtube later. \n\n**Abstract**:\nThe Critical Assessment of protein Function Annotation algorithms (CAFA) is an experiment designed to provide a large-scale assessment of computational methods dedicated to predicting protein function, using a time challenge. CAFA1-3 challenges have released a large number of approaches that can be be improved and used in CAFA5. I am going to review some of these approaches, concentrating on those developed by Hoehndorf lab as most successful.\n\nPS\n\nLiza Geraseva shared many remarkable notebooks for the competition (please upvote):\nhttps://www.kaggle.com/code/geraseva/keops-knn\nhttps://www.kaggle.com/code/geraseva/protbert-embedding\nhttps://www.kaggle.com/code/geraseva/deepgo\nAnd several others see: https://www.kaggle.com/geraseva/code\n\n========================================================\n\n## Previous webinar recordings: \n\nπΉ Video record of the webinar: https://youtu.be/h4zTjPzPud4\nπ Presentation: https://docs.google.com/presentation/d/1VrYMvJPlSrPd56pNXYm5PUylnMLEp9d0oMjcobOP47o/edit#slide=id.p\n\n========================================================\n\n## General announcement: \n\nThe competition is important for the research community. Thanks for the organizers and Kaggle ! \n\n- what is the best way to learn data science ? \n- participate in Kaggle competitions \n\nTo combine these two themes, my colleagues and me plan to organize some educational/research activity around that competition. The ideas are: to facilitate discussions, networking, knowledge sharing widely around that research topic and ideally to help to convert Kaggler's findings to some research papers, as well as to give some help to those who are less experienced but willing to participate. \nLet us emphasize - it is not intended to be private sharing / or even sharing of high scorable approaches, but rather sharing knowledge WIDELY around that research topic / that competition / competitions and data science in general. Whatever done will be publicly shared here on Kaggle forum. \n\nEverybody is welcome to participate ! \n\n- If you can find an opportunity to present something you find interesting on a webinar - your notebook, notebook by others, some research paper around, your experience on related task or anything related - you are very welcome ! Please contact me - and we will schedule it. Typically webinar is on zoom and after we put video records on youtube - see the previous activities here: https://www.youtube.com/playlist?list=PL1GnA8S_asBANHU95kOPOPEh3moPm5uhw \n- If you are a senior DS and can provide guidance for less experienced - please notify\n- If you are an expert in domain field - proteins - and have an opportunity to share knowledge - please notify\n- If you have some ideas , but do not have enough time to explore them - please notify - may be someone would find them interesting and would be able to try them. Typically many of us have more ideas, that we are able try. \n- any other ideas/suggestions welcome !\n\n**To start:** we plan an introductory webinar mainly for those who have not yet started. By Andrey Shevtsov ( who created and shared ESM2 embeddings: https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/406168 - please upvote ! )\n\nπ¨βπ¬ Andrey Shevtsov \"Introduction to CAFA5 Protein function prediction Kaggle competition\"\nβοΈ Thursday 22 June, 19.00 (CET time, e.g. Paris time) \nπΉ Video record of the webinar: https://youtu.be/h4zTjPzPud4\nπ Presentation: https://docs.google.com/presentation/d/1VrYMvJPlSrPd56pNXYm5PUylnMLEp9d0oMjcobOP47o/edit#slide=id.p\n\n \n[Add to google calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230622T170000Z%2F20230622T183000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=Andrey%20Shevtsov%20%22Introduction%20to%20CAFA5%20Protein%20function%20prediction%20Kaggle%20competition%22)\n\nThe **zoom link** will be posted here 10-15 minutes before the start. \n\nAbstract: \nThe competition to predict the functions of CAFA5 proteins continues at Kaggle (https://www.kaggle.com/competitions/cafa-5-protein-function-prediction). There are still two months to take part in it. Welcome to participate in our seminar at 22.06.2023 to listen to the introductory information on the competition.\nProteins are chains of amino acids (letters) folded into 3d structure. They perform many important functions in cells. The protein acquires its functions already in the folded state (in some cases - after additional modifications). But there is an opinion that the spatial arrangement of a protein molecule depends entirely on the original chain/sequence. This opinion is indirectly confirmed by the fact that there are methods for predicting the 3d structure of a protein by its sequence (like Alphafold/ESM family of models), as well as various encoders for performing downstream tasks. In this competition, the task is to predict protein properties by its sequence. The properties are presented in the form of directed acyclic graphs GO (gene ontology (http://geneontology.org/docs/ontology-documentation)). Interestingly, bioinformatic approaches are visible in the discussion.\n\n", "2325292": "## New webinar:\n\nπ¨βπ¬ Liza Geraseva \"A short review of methods used in CAFA challenges\"\nβοΈ Monday 3 July, 18.00 (CET time, e.g. Paris time)\n\n[Add to google calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230703T160000Z%2F20230703T183000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=Liza%20Geraseva%20%22A%20short%20review%20of%20methods%20used%20in%20CAFA%20challenges%22)\n\nThe zoom link will be posted here 10-15 minutes before the start. Video record will be on youtube later.\n\nAbstract:\nThe Critical Assessment of protein Function Annotation algorithms (CAFA) is an experiment designed to provide a large-scale assessment of computational methods dedicated to predicting protein function, using a time challenge. CAFA1-3 challenges have released a large number of approaches that can be be improved and used in CAFA5. I am going to review some of these approaches, concentrating on those developed by Hoehndorf lab as most successful.\n\nPS\n\nLiza Geraseva shared many remarkable notebooks for the competition (please upvote):\nhttps://www.kaggle.com/code/geraseva/keops-knn\nhttps://www.kaggle.com/code/geraseva/protbert-embedding\nhttps://www.kaggle.com/code/geraseva/deepgo\nAnd several others see: https://www.kaggle.com/geraseva/code\n\n", "2392430": "πΉ Video record: https://youtu.be/M8tqVF4Gyi0\nπ Presentation: https://t.me/sberlogabio/59283\n\nWelcome to the webinar on Friday: [John Mitchell ](https://www.kaggle.com/jbomitchell) who is expert both in bioinformatics and machine learning, and also experienced Kaggler and one of the top participants of CAFA5 , will share his experience on the past Kaggle competition \"Novoenzymes\" which in certain respects similar to CAFA5 challenge.\n\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)\n\nThe **zoom link** will be posted here 5 minutes before the start.\n\n[Add to google calendar\n](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230818T160000Z%2F20230818T173000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=%20John%20Mitchell%20%20%22Kaggle%20Competition%20Review%3A%20Novozymes%20Enzyme%20Stability%20Prediction%22)\n\n**Abstract:**\nKaggle's Novozymes Enzyme Stability Prediction was a challenging competition that rewarded expertise in bioinformatics as much as in Machine Learning. Two issues that made the competition particularly difficult were that the training data was rather different from the test data, and that there was no obvious or easy local validation protocol available.\nA wide variety of features and models proved valuable, while ensembling was generally considered essential. The nature of the competition lent itself to overfitting, and it was unsurprising that there was a large shake-up between Public LB and Private LB ranks.\nI discuss the experience gained from this competition and consider how the lessons learned might be applicable to other bioinformatic competitions.\n\nEverybody is welcome !\n\nVideo record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga", "2341826": "π¨βπ¬ Trifonova Sofya \"HMMER application to CAFA5.\"\nβοΈ Friday 14 July, 17.00 (CET time) \n\n[Add to google calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230714T140000Z%2F20230714T163000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=Trifonova%20Sofya%20%20%22HMMER%20application%20to%20CAFA5.%22)\n\nHMMER ( http://hmmer.org/ ) is a sensitive method for finding protein domains (https://en.wikipedia.org/wiki/Protein_domain) . Domains in proteins define it functional features, that is why using this method could help in better predictive ability of ML models in CAFA5.\n\nFor example, one may try to create 0-1 features enumerated by proteins domains: protein have particular domain (1) or not (0). And try to add them as predicators to the task. \n\nSofya created and shared several notebooks that one in particular: \nhttps://www.kaggle.com/code/sonyagrey/domains-encoding?scriptVersionId=136451490&cellId=9 (please upvote). \n\nThe **zoom link** will be posted here 10-15 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga \n", "2311366": "@serangu @simonveitner @zmcxjt Thank you for your kind words ! Would you be so kind to contact me al.chervov@gmail.com ", "2310179": "I am happy to share any insights I have gained so far in the field of automatic protein function prediction. I am also planning on presenting my preliminary work on using protein language models to predict function at ISMB/ECCB next month.", "2310111": "That's a great idea! Hope I can make it.", "2310088": "@alexandervc good idea! This is great for the community. The topic of the competition is difficult for Data Scientists. And the introductory lecture will help many!", "2313626": "i liked the meeting much today and i hope we can repeat it soon.", "2362609": "Presentation is attached\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"\nβοΈ Thursday 27 July \n\nπΉ Video: https://youtu.be/4W12qYMuMhk", "2361875": "Zoom link for today webinar:\nhttps://us02web.zoom.us/j/85437399900?pwd=QXZjcGNHaXpiYjRtUTlNMzZ1WXZDdz09\n\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"\nβοΈ Thursday 27 July, 19.00 (CET time)\n\nAdd to Google Calendar\n\nThe talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.\n\nThe zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga\n\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga\n\n", "2361238": "Reminder - today webinar:\n\nπ¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"\nβοΈ Thursday 27 July, 19.00 (CET time)\n\nAdd to Google Calendar\n\nThe talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.\n\nThe zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga\n\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga\n\n", "2357419": "π¨βπ¬ Henrik Nielsen,Vineet Thumuluri , JosΓ© Juan Almagro Armenteros \"DeepLoc 2.0: multi-label subcellular localization prediction using protein language models\"\nβοΈ Thursday 27 July, 19.00 (CET time)\n\n[Add to Google Calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230727T170000Z%2F20230727T193000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=Henrik%20Nielsen%2CVineet%20Thumuluri%20%2C%20Jos%C3%A9%20Juan%20Almagro%20Armenteros%20%20%22DeepLoc%202.0%3A%20multi-label%20subcellular%20localization%20prediction%20using%20protein%20language%20models%22)\n\nThe talk will be based on the recent paper with the same title (Nucleic Acids Res. 2022 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9252801/) ).\nThe prediction of protein subcellular localization is of great relevance for proteomics research. Here, we propose an update to the popular tool DeepLoc with multi-localization prediction and improvements in both performance and interpretability. For training and validation, we curate eukaryotic and human multi-location protein datasets with stringent homology partitioning and enriched with sorting signal information compiled from the literature. We achieve state-of-the-art performance in DeepLoc 2.0 by using a pre-trained protein language model. It has the further advantage that it uses sequence input rather than relying on slower protein profiles. We provide two means of better interpretability: an attention output along the sequence and highly accurate prediction of nine different types of protein sorting signals. We find that the attention output correlates well with the position of sorting signals. The webserver is available at services.healthtech.dtu.dk/service.php?DeepLoc-2.0.\n\nThe zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga\n\nπΉ Videos records are available : https://www.youtube.com/c/SciBerloga", "2349343": "πΉ Video: https://youtu.be/eGk2Y3zUe3Q \n\nEverybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.\n\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"\nβοΈ Monday 17 July, 17.00 (CET time) \n\n\n Sequence-structure based bioinformatics approach is a new direction of bioinformatics which is based on representation of sequences as their structural formula and analysis of structure-property relationships using molecular descriptors and machine learning algorithms. Some cases using of this approach will be discussed during webinar. \n\nAbout the reporter: Prof. A. Lagunin (https://scholar.google.ru/citations?user=qaSuEUkAAAAJ&hl ) authored numerous publications in the field of bioinformatics.", "2348350": "https://us02web.zoom.us/j/89020453099?pwd=aXRGUkRwUXBSaHozSk5QU1lsV0NPQT09\n\nEverybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.\n\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"\nβοΈ Monday 17 July, 17.00 (CET time) \n", "2345869": "Everybody is welcome to a research/educational webinar. The methods discussed might be useful for CAFA5 participants, although the webinar is NOT specific to CAFA5 challenge. However molecular descriptors developed by the authors see e.g. https://europepmc.org/article/med/29080875 might be used for feature generation - although the publication is devoted to the different task, might be some modification to CAFA5 challenge is possible - hopefully it might be more clear after the webinar.\n\nπ¨βπ¬ Prof. Alexey Lagunin \"Sequence-structure based bioinformatics.\"\nβοΈ Monday 17 July, 17.00 (CET time)\n\n[Add to Google Calendar](https://calendar.google.com/calendar/render?action=TEMPLATE&dates=20230717T150000Z%2F20230717T173000Z&details=Zoom%20link%20will%20be%20available%20at%20Kaggle%20forum%20shortly%20before%20the%20start.&text=Prof.%20A.Lagunin%20%22Sequence-structure%20based%20bioinformatics.%22)\n\nSequence-structure based bioinformatics approach is a new direction of bioinformatics which is based on representation of sequences as their structural formula and analysis of structure-property relationships using molecular descriptors and machine learning algorithms. Some cases using of this approach will be discussed during webinar.\n\nAbout the reporter: Prof. A. Lagunin (https://scholar.google.ru/citations?user=qaSuEUkAAAAJ&hl ) authored numerous publications in the field of bioinformatics.\n\nThe zoom link will be posted here 5 minutes before the start. Video record will be on youtube later on the channel: https://www.youtube.com/@SciBerloga", "2344545": "https://us02web.zoom.us/j/81847029643?pwd=bkkvYUhESmViemFTdExRMDdyMDYzQT09\n\nπ¨βπ¬ Trifonova Sofya \"HMMER application to CAFA5.\"\nβοΈ Friday 14 July, 17.00 (CET time) \n", "2344429": "> π¨βπ¬ Trifonova Sofya \"HMMER application to CAFA5.\"\n> βοΈ Friday 14 July, 17.00 (CET time)\n\n\nLooks like the Google calendar invitation for July 14 is for 16:00 CET? Anyhow, I'm looking forward to it.\n\nIddo", "2334444": "Zoom link: \n\nhttps://us02web.zoom.us/j/88968811599?pwd=YitNc0RaeFVRSmNqemxoQ2dKWnJZZz09\n\nπ¨βπ¬ Marina Pak \"ProDDG (https://ivankovlab.ru/proddg)- database of data on protein stability change upon mutation (ΞΞG) and predictors of mutation effect.\"\nβοΈ Friday 7 July, 19.00 (CET time)\n\n", "2329313": "Thank you for recording \n\nπΉ Video: https://youtu.be/hpYfLLqyo9Y\nβοΈ Abstract: https://t.me/sberlogabig/278", "2397561": "πΉ Video: https://youtu.be/M8tqVF4Gyi0\n\n\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)\n\n", "2396945": "Zoom link:\nhttps://us02web.zoom.us/j/84399146917?pwd=TitKVm5NbS9tWTNrcm96SjY1dnFhZz09\n\nWelcome to the webinar on Friday: John Mitchell who is expert both in bioinformatics and machine learning, and also experienced Kaggler and one of the top participants of CAFA5 , will share his experience on the past Kaggle competition \"Novoenzymes\" which in certain respects similar to CAFA5 challenge.\n\nπ¨βπ¬ John Mitchell \"Kaggle Competition Review: Novozymes Enzyme Stability Prediction\"\nβοΈ Friday 18 August, 18.00 (CET time, e.g. Paris time)\n\n", "2348314": "Zoom meetings will start in half an hour?", "2313413": "" } }