shamsaraj

shamsaraj predicts inhibitory activity of soluble epoxide hydrolase (sEH) inhibitors using a Random Forest QSAR model trained on simple fragmental descriptors.

Key Features:

• Machine Learning Approach: Uses the Random Forest algorithm to build a QSAR model for sEH inhibition prediction.
• Fragmental Descriptors: Represents compounds with simple fragmental descriptors derived from molecular fragments.
• Model Interpretability: Applies Treeinterpreter (Python) and LIME to attribute model predictions to fragmental features.
• Performance Metrics: Model metrics reported as R² = 0.831, Q² = 0.565, RMSE = 0.552, and R² pred = 0.595.
• External Validation: Validated on two external test sets with Spearman's rank correlations of 0.872 and 0.673, noting the second set is more diverse than the training data.

Scientific Applications:

• Virtual Screening: Ranks and enriches potential sEH inhibitors from large and diverse chemical libraries.
• Medicinal Chemistry Guidance: Identifies important fragmental descriptors consistent with crystallographic data to inform design of new sEH inhibitors.

Methodology:

A Random Forest QSAR model was trained on known sEH inhibitors represented by simple fragmental descriptors, and Treeinterpreter and LIME were applied to elucidate fragment contributions and compare them with crystallographic structural data.