A Robust Drug-Target Interaction Prediction Framework with Capsule Network and Transfer Learning.

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Drug-target interactions (DTIs) are vital for drug design and discovery.
• Current computational methods for DTI prediction face limitations due to insufficient negative datasets, inaccurate feature representation, and ineffective classifiers.

Purpose of the Study:

• To develop a robust computational framework for accurate DTI prediction.
• To address limitations in existing DTI prediction methodologies.

Main Methods:

• Proposed CapBM-DTI, a capsule network-based framework.
• Utilized pre-trained bidirectional encoder representations from transformers (BERT) for protein sequence feature extraction.
• Employed message-passing neural networks (MPNN) for compound graph feature extraction.
• Established two experimentally validated datasets for training and evaluation.

Main Results:

• CapBM-DTI demonstrated robust performance and strong generalization ability across various DTI datasets.
• The model outperformed state-of-the-art methods in predicting DTIs.
• A case study highlighted the model's applicability in virtual screening for COVID-19 drug discovery.

Conclusions:

• The CapBM-DTI framework offers an accurate and robust approach to identifying drug-target interactions.
• The study provides a valuable tool for accelerating drug discovery and virtual screening processes.