Machine learning-based classification models for non-covalent Bruton's tyrosine kinase inhibitors: predictive ability...

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Bruton's tyrosine kinase (BTK) is a key target in treating various cancers and autoimmune diseases.
• Developing selective non-covalent BTK inhibitors requires efficient prediction of their bioactivity.
• Machine learning offers a powerful approach for predicting molecular properties and guiding drug design.

Purpose of the Study:

• To develop and validate machine learning models for predicting the bioactivity of non-covalent BTK inhibitors.
• To provide interpretable explanations for model predictions using explainable AI techniques.
• To facilitate the design of novel BTK inhibitors with enhanced efficacy and selectivity.

Main Methods:

• Collected a dataset of 3895 non-covalent BTK inhibitors from Reaxys and ChEMBL databases.
• Utilized MACCS and Morgan molecular fingerprints for feature representation.
• Trained and evaluated traditional (DT, RF, SVM, XGBoost) and deep learning (DNN) classification models.
• Applied SHAP for model interpretability and K-means/hierarchical clustering for structural visualization.

Main Results:

• The best model (XGBoost with MACCS fingerprints) achieved 94.1% accuracy and an MCC of 0.75.
• SHAP analysis successfully decomposed predictions, highlighting key molecular features contributing to bioactivity.
• Clustering analysis revealed distinct groups of inhibitors, consistent with crystal structure interactions.

Conclusions:

• Machine learning models demonstrate high predictive performance for non-covalent BTK inhibitor bioactivity.
• Explainable AI (SHAP) enhances model transparency and aids in understanding structure-activity relationships.
• The developed models and insights are valuable for accelerating the discovery and optimization of BTK inhibitors.