GoMA-DTA: A Gene Ontology-Guided Multimodal Attention Fusion Model for Drug-Target Affinity Prediction

Area of Science:

• Computational Biology
• Drug Discovery
• Bioinformatics

Background:

• Accurate drug-target affinity (DTA) prediction is crucial for efficient drug discovery.
• Existing methods often rely on sequence patterns, neglecting high-level biological functions.
• Pretrained protein language models show promise but require enhanced functional integration.

Purpose of the Study:

• To develop GoMA-DTA, a framework integrating gene ontology (GO) functional annotations with protein semantic features for improved DTA prediction.
• To introduce adaptive semantic filtering by recalibrating protein embeddings using functional semantics.
• To enhance drug representation by combining semantic and structural features for synergistic interaction with protein features.

Main Methods:

• GoMA-DTA utilizes a channelwise gating mechanism for adaptive semantic filtering of protein embeddings (ESM-2).
• Drug representations are generated using Molformer for semantics and TransConv for structure.
• A parallel synergistic architecture with cross-attention and Mamba modules facilitates cross-modal alignment and long-range dependency modeling.

Main Results:

• GoMA-DTA significantly outperforms state-of-the-art models on PDBBind, BindingDB, and ChEMBL benchmarks.
• The model demonstrates superior screening power, validated on the CASF-2016 dataset.
• Virtual screening against SARS-CoV-2 Spike protein showed practical utility, supported by experimental evidence.

Conclusions:

• GoMA-DTA offers a robust and biologically reliable tool for drug discovery by integrating functional genomics.
• The framework achieves precise cross-modal alignment and efficient modeling of complex interactions.
• GoMA-DTA represents a significant advancement in DTA prediction, accelerating the identification of potential drug candidates.