Meta-Modeling with Drug Discovery Stack Regressor for Drug Discovery: An Explainable AI Perspective.

Area of Science:

• Artificial Intelligence in Drug Discovery
• Computational Chemistry
• Machine Learning for Pharmacology

Background:

• Drug discovery is challenged by complex datasets, long timelines, and inaccurate predictions of drug-target interactions.
• These challenges impede timely therapeutic development, particularly during global health crises like COVID-19.
• This study addresses these issues by integrating ensemble machine learning with explainable artificial intelligence (XAI).

Purpose of the Study:

• To enhance the predictive accuracy and transparency of AI models in drug discovery.
• To leverage ensemble methods and XAI for more robust and interpretable drug-target interaction predictions.
• To provide chemically meaningful insights for informed molecular design.

Main Methods:

• Trained three regression models (Random Forest, Support Vector Regression, Multi-Layer Perceptron) on 104 COVID-19 compounds.
• Implemented ensemble strategies: Voting Regressors and Stacking Regressors.
• Utilized SHAP (SHapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations) for feature importance analysis.

Main Results:

• The Drug Discovery Stack Regressor model demonstrated superior performance with MSE of 0.18 and R² of 0.88.
• SHAP and LIME identified EffectiveRotorCount3D and YStericQuadrupole3D as key molecular descriptors.
• These features relate to molecular flexibility and steric effects crucial for drug activity.

Conclusions:

• Combining ensemble modeling with explainability significantly improves prediction robustness and interpretability in drug discovery.
• SHAP and LIME integration provides chemically relevant insights, supporting rational molecular design and increasing model transparency.
• Explainable ensemble models enhance the reliability and applicability of AI in drug discovery, offering scalable solutions for therapeutic development.