TransGAT-DTA: A multi-task framework for drug-target affinity prediction and conditional molecule generation.

Area of Science:

• Computational chemistry
• Machine learning in drug discovery
• Bioinformatics

Background:

• Drug discovery is hindered by the time and cost of identifying molecules that interact with target proteins.
• Existing machine learning models often operate in single-task settings, limiting their ability to address both affinity prediction and molecule generation concurrently.

Purpose of the Study:

• To develop a multi-task learning framework, TransGAT-DTA, capable of simultaneous drug-target affinity prediction and targeted molecule generation.
• To overcome the limitations of single-task models in computational drug discovery.

Main Methods:

• Utilized a multi-task learning framework integrating a shared graph-Transformer for molecular features and gated CNNs for protein sequences.
• Implemented a learnable alignment module for cross-modal feature integration.
• Employed a dynamic gradient coordination mechanism and conditional control attention for balanced optimization and guided molecule generation.

Main Results:

• TransGAT-DTA demonstrated reduced mean squared error in affinity prediction compared to single-task models.
• Successfully generated high-quality, target-specific molecules.
• The end-to-end design mitigated error accumulation, offering robust bidirectional target-molecule optimization.

Conclusions:

• TransGAT-DTA provides an effective framework for simultaneous drug-target affinity prediction and molecule generation.
• The model enhances efficiency and accuracy in drug discovery pipelines.
• Establishes a foundation for advancements in multi-target drug discovery.