{ "fetched_at": "2026-10-19T16:44:46Z", "source_url": "https://www.kaggle.com/competitions/euos-slas/data", "pages": [ { "name": "data-description", "content": "This challenge data includes 101,017 data points in total. Challenge participants are expected to use the training data to build a prediction model that can predict the solubility class in the test set. The predictors can be generated from embedding the SMILES and/or collected from public resources, such as https://ecbd.eu/.\nAn example predictor matrix generated using SMILES is shown in the example_predictors.csv. \n\nSolubility labels are provided for 70% of the data. The rest 30% of the data points are equally spilt for the public leaderboard and final evaluation. All splitings are conducted using stratified random sampling with the solubility category as the stratification factor.\n\n## Files\n\n* **train.csv** - the training set\n* **test.csv** - the test set\n* **submission_template.csv** - a sample submission file in the correct format\n* **example_predictors** - an example predictors metrics generated from SMILES id, Note! Challenge participants are NOT REQUIRED to use this file.\n\n## Columns\n\n* `Id` - ID column of the compoud\n* `smiles` - [Simplified Molecular Input Line Entry System](https://dev.drugbank.com/guides/terms/smiles)\n* `sol_category` Solubility categories from the experiments.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Description", "content": "### WHAT DRIVES US\n\nKnowledge about the chemical properties of small molecules is an essential feature to understand their behaviour within their environment. Scientists in academia and big pharma synthesize and screen millions of molecules to discover new chemical probes and drugs. Such molecules can enlarge our understanding of biology and provide treatments for severe diseases that cause millions of deaths every year. Many small molecules are not well characterized, and with the sheer number of available or synthesizable compounds, it seems unlikely that this will ever happen. Therefore, ***the meaningful prediction of chemical properties of novel compounds will provide an essential tool to speed up research and allows us to focus more on the complex and often even less understood biology.***\n\n### WHAT WE EXPECT FROM YOU\n\nEU-OPENSCREEN ERIC and SLAS challenge you to develop a reliable algorithm that can predict the solubility of a small molecule, an essential feature of all biologically active compounds. \n\n#### WHAT WE PROVIDE\n\nEU-OPENSCREEN ERIC provides a high-quality data set of experimentally measured aqueous solubility of about 100,000 small molecules which was produced at an EU-OPENSCREEN ERIC high throughput screening partner site. 70,000 of these molecules will be available for download on Kaggle, and the residual 30,000 compounds will be withheld for prediction. \n\n### Protocol used to measure solubility\n\nCompounds were dispensed (50 nL from a 10 mM solution in DMSO) to the assay plates (Greiner, Ref: 781801) by employing an Echo550 dispenser. Amiodarone and Phenytoin were employed as assay controls\n50 µL of PBS (pH=7.4) were added to each well to achieve a 10 µM concentration with a 0.1% of DMSO.\nPlates were sealed and incubated at 37°C under gentle stirring for 2 h.\nAfter the incubation plates were centrifuged for 20 sec at 200xg and the nephelometry was detected in a Nephelostar reader (BMG Labtech).\nThe entire screening was run in duplicates (N=2), but the duplicates were not in the same plates. Thus, the screening was carried out by assaying two identical plates with the compounds in the same location. This means that there were 352 compounds per plate in two identical plates for calculating the average and SD. The other wells were employed for blank measurements and controls. Data was analyzed by subtracting the nephelometry values observed in wells containing 0.1% DMSO. The solubility threshold for classifying compounds as high, medium and low solubility was fixed based on the average nephelometry values observed for phenytoin (high solubility) and amiodarone (medium-low solubility). Thus, the thresholds were fixed as follows:\nHigh solubility: Nephelometry values lower than 50000\nMedium solubility: Nephelometry values between 50000 and 100000\nLow solubility: Nephelometry values higher than 100000\n\n### Evaluation of models\n\nResults will be evaluated using the Kappa coefficient calculated on the independent test set. The winner will receive the opportunity to present their winning algorithm in an SLAS journal, SLAS Discovery, and will be given the opportunity to present his/her research at the 2023 European SLAS conference in Brussels. \n\n### ACKNOWLEDGMENTS\n\nWe thank Prof. Dr. Mabel Loza and Dr. Jose Brea of the BioFarma Research group from the University of Santiago de Compostela for the preparation of the high-quality solubility data set.\n\nWe thank Dr. Andrea Zaliani from the Fraunhofer Institute for Translational Medicine and Pharmacology, Prof. Dr. Jing Tang and Wenyu Wang from the University of Helsinki for their expert input on data curation, scoring algorithms and role as competition judges.\n\nWe thank Dr. Julio Martin, Dr. Christian Parker, Jill Hronek, Kathy Daniels and Dr. Hannah Rosen from the SLAS and Dr. Wolfgang Fecke and Dr. Robert Harmel from EU-OPENSCREEN ERIC for organizing the competition.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Evaluation", "content": "Each participating team can have at most one submission per day in the public leaderboard phase and only ONE submission in the final phase. The performance in the public leaderboard phase is expected to help participants improve their model. The ranking of the participating teams is decided using a final validation set independent of the test data used in the public and private leaderboard estimation.\n\nThe primary metric for this competition is [Quadratic Kappa Metric](https://www.kaggle.com/code/aroraaman/quadratic-kappa-metric-explained-in-5-simple-steps/notebook). \n\nFor the final evaluation, in addition to the point estimates calculated on the final validation set (scores calculated by Kaggle), we will also apply a bootstrapping resampling to identify statistically tied top-performing teams. To this end, we will draw 1000 bootstrapped samples (with replacement) from the final validation set to generate the distribution of each model' Kappa score. A Bayer factor, Km will be calculated for each team's final model, with the model showing the highest point estimation as the reference. \n\nAs a consequence, participants that want to officially win the competition and the prize need to share their algorithms. However, sharing algorithms is not required to test your skills and take part in the competition. \n\n\n\n## Submission Format\n\n**For every compound in the test dataset**, submission files should contain two columns: `Id` and `Pred`. \n\n\nThe file should contain a header and have the following format:\n```\nId,Pred\n1,1\n```", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "rules", "content": "##One account per participant\nYou cannot sign up to Kaggle from multiple accounts and therefore you cannot submit from multiple accounts.\n\n##No private sharing outside teams\nPrivately sharing code or data outside of teams is not permitted. It's okay to share code if made available to all participants on the forums.\n\n##Submission Limits\nYou may submit a maximum of 1 entry per day during the public leaderboard phase. \n\n##Open Science\nParticipants, whether individual or part of a team, is not allowed to use the OPENSCREEN dataset for commercial purpose without permission. The winning team is obliged to acknowledge the OPENSCREEN consortium and share their code and results if asked.\n \nYou may select up to 1 final submission for judging.\n\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Prize", "content": "- **Publication of a technical brief or protocol article with the winning algorithm in an SLAS journal without APC.**\nhttps://www.slas.org/publications/slas-discovery/\n\n- **A podium talk at SLAS Europe 2023 with registration and travel costs covered.**\nhttps://www.slas.org/events-calendar/slas-europe-2023-conference-and-exhibition/\n\n- **One-year free SLAS membership.**\nhttps://www.slas.org/\n\n\n\n\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Timeline", "content": "- **19th September 2022**: Competition Start\n- **31st December 2022**: Final Submission Deadline\n- **31st January 2023**: Announcement of the Winners\n- **22nd-26th May 2023**: Presentation of the Winner at SLAS Science Set in Motion Brussels", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Instructions For Data Submission ", "content": "## Instructions for Data Submission\nParticipants should submit their prediction in the same format as shown in the \n`submimssion_template.csv` under `Data` section. \nNOTE! The predicted values (Predicted) can only be 2, 1, or 0, representing whether the compound is classified into HIGH, MEDIUM, or LOW solubility classs.\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Winners", "content": "1. a.kopp.chem\n2. Beardy Polonium\n3. Mr. Maniac", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false } ] }