MOLECULE: Molecular-dynamics and Optimized deep Learning for Entropy-regularized Classification and Uncertainty-aware...

Area of Science:

• Computational chemistry
• Pharmacology
• Biophysics

Background:

• Machine learning (ML) and deep learning (DL) are advancing drug discovery.
• Integrating structure-based data enhances ML model predictions.
• Previous work showed ML models can classify kinase ligands using molecular dynamics (MD) data.

Purpose of the Study:

• To develop a novel dual-modal deep neural network classifier for predicting compound mode of action.
• To process dynamical and structural data separately and efficiently.
• To leverage a diverse kinase dataset with 280 experimentally resolved structures.

Main Methods:

• Curated a diverse kinase dataset (280 structures).
• Developed a dual-modal deep neural network classifier.
• Trained and evaluated the model on dynamical and structural data.

Main Results:

• Achieved robust classification performance in predicting compound mode of action.
• Demonstrated effective uncertainty handling.
• Highlighted the importance of incorporating protein dynamics data.
• Maintained high performance with imputed dynamics data.

Conclusions:

• The developed dual-modal deep neural network effectively predicts compound mode of action.
• Protein dynamics data is critical for accurate predictions.
• The method allows rapid compound screening and prioritization without extensive simulations.