I understand that for model building we can use published tools like BLAST, protein language models, and AlphaGO2 that predicts various biochemical aspects of proteins. My question is does this extend to models that directly predict protein GO terms if they are publicly available? Can we use the results of such models when creating our ensemble predictions for this competition? One example is the recently published ProteInfer (https://github.com/google-research/proteinfer)
", "messages": [ { "id": "2311161", "postDate": "06/21/2023 01:10:32", "content": "I understand that for model building we can use published tools like BLAST, protein language models, and AlphaGO2 that predicts various biochemical aspects of proteins. My question is does this extend to models that directly predict protein GO terms if they are publicly available? Can we use the results of such models when creating our ensemble predictions for this competition? One example is the recently published ProteInfer (https://github.com/google-research/proteinfer)
", "rawMarkdown": "I understand that for model building we can use published tools like BLAST, protein language models, and AlphaGO2 that predicts various biochemical aspects of proteins. My question is does this extend to models that directly predict protein GO terms if they are publicly available? Can we use the results of such models when creating our ensemble predictions for this competition? One example is the recently published ProteInfer (https://github.com/google-research/proteinfer)", "votes": null }, { "id": "2311170", "postDate": "06/21/2023 01:37:56", "content": "\n\nquestion is does this extend to models that directly predict protein GO terms if they are publicly available? Can we use the results of such models when creating our ensemble predictions for this competition?
\n
In one word: yes. Be advised of the Winner License clause in the rules though.
", "rawMarkdown": "> question is does this extend to models that directly predict protein GO terms if they are publicly available? Can we use the results of such models when creating our ensemble predictions for this competition?\n\nIn one word: yes. Be advised of the Winner License clause in the rules though.", "votes": null }, { "id": "2311186", "postDate": "06/21/2023 02:03:11", "content": "thank you, that is good to know. What if the model we use depends upon predictions of models developed by others who have not open sourced their code?
", "rawMarkdown": "thank you, that is good to know. What if the model we use depends upon predictions of models developed by others who have not open sourced their code?", "votes": null }, { "id": "2311434", "postDate": "06/21/2023 07:01:52", "content": "I suspect no. I asked the guys behind SPROF-GO if they could add a licence to the code before posting predictions.
", "rawMarkdown": "I suspect no. I asked the guys behind SPROF-GO if they could add a licence to the code before posting predictions.", "votes": null }, { "id": "2311975", "postDate": "06/21/2023 15:23:02", "content": "I hope that is true but from what I can see many or most of the top model predictors for CAFA3 are from groups that did not open source their solutions. Hopefully there will be some additional clarity on this. For example Ankh and Prot-T5 are not strictly open sourced (AFAIK) but many of our submissions are based upon the pre-trained weights/embeddings which anyone can download. Likewise there are webtools that predict GO terms which are publicly available but the underlying models are not open sourced.
", "rawMarkdown": "I hope that is true but from what I can see many or most of the top model predictors for CAFA3 are from groups that did not open source their solutions. Hopefully there will be some additional clarity on this. For example Ankh and Prot-T5 are not strictly open sourced (AFAIK) but many of our submissions are based upon the pre-trained weights/embeddings which anyone can download. Likewise there are webtools that predict GO terms which are publicly available but the underlying models are not open sourced.", "votes": null } ], "comments": [ { "id": 2311170, "author_name": "iddux1", "author_url": "", "post_date": "06/21/2023 01:37:56", "content": "\n\nquestion is does this extend to models that directly predict protein GO terms if they are publicly available? Can we use the results of such models when creating our ensemble predictions for this competition?
\n
In one word: yes. Be advised of the Winner License clause in the rules though.
", "votes": null, "replies": [ { "id": 2311186, "author_name": "zmcxjt", "author_url": "", "post_date": "06/21/2023 02:03:11", "content": "thank you, that is good to know. What if the model we use depends upon predictions of models developed by others who have not open sourced their code?
", "votes": null, "replies": [ { "id": 2311434, "author_name": "mtinti", "author_url": "", "post_date": "06/21/2023 07:01:52", "content": "I suspect no. I asked the guys behind SPROF-GO if they could add a licence to the code before posting predictions.
", "votes": null, "replies": [ { "id": 2311975, "author_name": "zmcxjt", "author_url": "", "post_date": "06/21/2023 15:23:02", "content": "I hope that is true but from what I can see many or most of the top model predictors for CAFA3 are from groups that did not open source their solutions. Hopefully there will be some additional clarity on this. For example Ankh and Prot-T5 are not strictly open sourced (AFAIK) but many of our submissions are based upon the pre-trained weights/embeddings which anyone can download. Likewise there are webtools that predict GO terms which are publicly available but the underlying models are not open sourced.
", "votes": null, "replies": [] } ] } ] } ] } ], "raw_markdown_by_id": { "2311161": "I understand that for model building we can use published tools like BLAST, protein language models, and AlphaGO2 that predicts various biochemical aspects of proteins. My question is does this extend to models that directly predict protein GO terms if they are publicly available? Can we use the results of such models when creating our ensemble predictions for this competition? One example is the recently published ProteInfer (https://github.com/google-research/proteinfer)", "2311170": "> question is does this extend to models that directly predict protein GO terms if they are publicly available? Can we use the results of such models when creating our ensemble predictions for this competition?\n\nIn one word: yes. Be advised of the Winner License clause in the rules though.", "2311186": "thank you, that is good to know. What if the model we use depends upon predictions of models developed by others who have not open sourced their code?", "2311434": "I suspect no. I asked the guys behind SPROF-GO if they could add a licence to the code before posting predictions.", "2311975": "I hope that is true but from what I can see many or most of the top model predictors for CAFA3 are from groups that did not open source their solutions. Hopefully there will be some additional clarity on this. For example Ankh and Prot-T5 are not strictly open sourced (AFAIK) but many of our submissions are based upon the pre-trained weights/embeddings which anyone can download. Likewise there are webtools that predict GO terms which are publicly available but the underlying models are not open sourced." }, "source": "meta" }