Accelerating Discovery of Leukemia Inhibitors Using AI-Driven Quantitative Structure-Activity Relationship: Algorithm...

Area of Science:

• Computational chemistry and cheminformatics
• Oncology and drug discovery
• Machine learning applications in medicinal chemistry

Background:

• Leukemia treatment faces challenges due to limited potency and selectivity of current inhibitors.
• Traditional drug discovery methods struggle to navigate complex chemical spaces for novel inhibitors.
• Machine learning (ML)-enhanced quantitative structure-activity relationship (QSAR) modeling offers a computational strategy to optimize drug candidates.

Purpose of the Study:

• To develop and validate an integrated ML-enhanced QSAR workflow for rational design of Thiadiazolidinone (TDZD) analogs.
• To predict and optimize TDZD analogs with improved anti-leukemia activity.
• To identify key molecular determinants of potency and guide future inhibitor optimization.

Main Methods:

• Analysis of 35 TDZD derivatives with confirmed anti-leukemia activity.
• Calculation of 220 molecular descriptors (1D-4D) using Schrödinger MAESTRO.
• Training and testing of 17 ML models (e.g., Random Forests, XGBoost, Neural Networks) using stratified sampling and 5-fold cross-validation.
• Performance evaluation using 12 metrics, including MSE, R², and SHAP values, with hyperparameter tuning.

Main Results:

• Ensemble methods, particularly LightGBM and Random Forest, demonstrated superior predictive performance (LightGBM: R² = 0.971).
• Modest training-to-test performance degradation indicated genuine pattern learning.
• Key features influencing anti-leukemia activity included molecular shape, polar surface area, polarizability, partition coefficient, and hydrogen bonding capacity.

Conclusions:

• Integrated ML and QSAR modeling effectively analyze structure-activity relationships for TDZD analogs.
• Ensemble methods show high internal validation, but external validation and experimental testing are crucial.
• Identified molecular features provide a foundation for future validation and optimization of anti-leukemia inhibitors.