Outline and background for the EU-OS solubility prediction challenge

Area of Science:

• Chemoinformatics
• Machine Learning
• Drug Discovery

Background:

• The European Organisation for the Safety and Health at Work (EU-OS) released a substantial dataset of compound solubility.
• This dataset comprises over 100,000 compounds from proprietary screening collections.
• There is a growing need for accurate compound solubility prediction in scientific research.

Purpose of the Study:

• To foster the development of predictive classification models for compound solubility.
• To leverage a large, publicly available dataset for machine learning model training.
• To address the challenge of predicting compound solubility through community-driven efforts.

Main Methods:

• A joint competition was organized by EU-OS and the Society for Laboratory Automation and Screening (SLAS).
• The competition utilized the Kaggle data science platform for its versatile and open-source nature.
• Chemoinformatics and machine learning experts were invited to develop predictive models.

Main Results:

• A competition was successfully conducted on the Kaggle platform.
• The competition aimed to identify the most predictive classification model for compound solubility.
• Considerations on the competition's results and encountered challenges are presented.

Conclusions:

• The competition facilitated the development of advanced machine learning models for solubility prediction.
• The initiative utilized the largest publicly available training set for solubility prediction.
• The findings contribute to the broader understanding of predicting compound solubility.