A Scalable Structure-Aware Multimodal Architecture for Accurate Drug-Target Affinity Prediction

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate drug-target binding affinity (DTA) prediction is crucial for virtual screening.
• Existing methods like sequence-based approaches lack spatial details, while structure-based methods are computationally intensive.
• There is a need for scalable and accurate DTA prediction methods that balance performance and efficiency.

Purpose of the Study:

• To introduce StructuraDTA, a novel multimodal framework for DTA prediction.
• To address the limitations of current computational methods by employing an implicit structure modeling strategy.
• To develop a scalable and accurate solution for accelerating genome-scale drug discovery.

Main Methods:

• Encoding drug molecular graphs using Graph Isomorphism Networks (GINs) for topological feature extraction.
• Optimizing protein representations by integrating probabilistic structural priors into a pretrained language model, simulating conformational flexibility without explicit 3D data.
• Utilizing a bidirectional cross-attention mechanism for dynamic alignment of heterogeneous feature modalities.

Main Results:

• StructuraDTA consistently outperforms state-of-the-art methods on Davis and KIBA benchmark datasets.
• The model demonstrates strong robustness in cold-start scenarios, accurately predicting affinities for novel drugs and targets.
• Achieved high predictive performance comparable to structure-based models with the inference efficiency of sequence-based methods.

Conclusions:

• StructuraDTA offers an accurate and scalable solution for drug-target binding affinity prediction.
• The implicit structure modeling approach effectively captures essential spatial and topological information.
• This framework has the potential to significantly accelerate genome-scale drug discovery research.