ChemInformatics Model Explorer (CIME): exploratory analysis of chemical model explanations

Area of Science:

• Computational Chemistry
• Cheminformatics
• Artificial Intelligence in Drug Discovery

Background:

• Machine learning enhances efficiency in small molecule research by predicting properties and bioactivity.
• Explainable AI (XAI) offers insights into model reasoning, crucial for understanding compound-property relationships.
• Current tools lack interactive visualization for interdisciplinary collaboration on ML model interpretability in chemistry.

Purpose of the Study:

• To introduce CIME (ChemInformatics Model Explorer), an interactive web-based system.
• To facilitate interdisciplinary collaboration by enabling visualization of machine learning model explanations.
• To address the need for tools that help chemists and data scientists interpret ML models in small molecule research.

Main Methods:

• Development of CIME, a model-agnostic, web-based interactive system.
• CIME allows inspection of chemical datasets and visualization of model explanations.
• The system supports comparison of interpretability techniques and exploration of compound subgroups.

Main Results:

• CIME provides an interactive platform for exploring and understanding machine learning models in cheminformatics.
• Users can visualize how compound substructures contribute to predicted properties.
• The tool supports collaborative analysis and decision-making in small molecule research.

Conclusions:

• CIME bridges the gap between machine learning interpretability and practical application in small molecule research.
• The tool enhances interdisciplinary collaboration between chemists and data scientists.
• CIME offers a valuable resource for understanding and utilizing complex predictive models in drug discovery.