CIME

CIME enables model-agnostic visualization and analysis of machine-learning explanations for small-molecule datasets to support interpretation of bioactivity and pharmacokinetic predictions.

Key Features:

• Model-agnostic analysis: Applies to predictions from arbitrary machine learning models without requiring model-specific integration.
• Visualization of model explanations: Visualizes how chemical substructures contribute to predicted properties at the compound level.
• Interpretability technique comparison: Compares multiple interpretability techniques to contrast model reasoning across methods.
• Chemical dataset inspection: Enables granular inspection of chemical datasets and compound-level predictions.
• Subgroup and cluster exploration: Supports exploration of compound subgroups to identify patterns or clusters within datasets.

Scientific Applications:

• Small-molecule research: Facilitates interpretation of machine-learning results in small-molecule research.
• Pharmacokinetics prediction: Assists interpretation and prioritization of pharmacokinetic property predictions.
• Bioactivity prediction: Aids interpretation and prioritization of bioactivity prediction models.
• Compound design optimization: Informs refinement of compound design and optimization of chemical properties by revealing substructure contributions.
• Explainable AI for molecular ML: Supports explainable artificial intelligence workflows for molecular machine learning models.

Methodology:

Model-agnostic analysis of machine-learning outputs; visualization of substructure-level explanations; comparison of multiple interpretability techniques; subgroup/cluster exploration and granular inspection of chemical datasets.