Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Structure-Aware Contrastive Learning with Fine-Grained Binding Representations for Drug Discovery

Jing Lan, Hexiao Ding, Hongzhao Chen, Yufeng Jiang, Nga-Chun Ng, Gwing Kei Yip, Gerald W. Y. Cheng, Yunlin Mao, Jing Cai, Liang-ting Lin, Jung Sun Yoo

TL;DR

The paper tackles scalable drug–target interaction (DTI) prediction by embedding structural priors into sequence-based representations. It introduces a structure-aware protein vocabulary, SELFIES-based drug encoding, a patch-based attention trunk, and a contrastive cross-modal objective paired with a bilinear attention module to model interactions. The approach achieves state-of-the-art results on the Human and BioSNAP DTI benchmarks, remains competitive on BindingDB, and delivers superior virtual screening performance on LIT-PCBA (AUROC 68.16% and BEDROC 13.35%). Ablation and visualization analyses confirm the importance of learned aggregation, contrastive alignment, and interpretable attention over ligand–residue contacts, highlighting the method’s robustness and interpretability for scalable, structure-informed DTI pre-screening.

Abstract

Accurate identification of drug-target interactions (DTI) remains a central challenge in computational pharmacology, where sequence-based methods offer scalability. This work introduces a sequence-based drug-target interaction framework that integrates structural priors into protein representations while maintaining high-throughput screening capability. Evaluated across multiple benchmarks, the model achieves state-of-the-art performance on Human and BioSNAP datasets and remains competitive on BindingDB. In virtual screening tasks, it surpasses prior methods on LIT-PCBA, yielding substantial gains in AUROC and BEDROC. Ablation studies confirm the critical role of learned aggregation, bilinear attention, and contrastive alignment in enhancing predictive robustness. Embedding visualizations reveal improved spatial correspondence with known binding pockets and highlight interpretable attention patterns over ligand-residue contacts. These results validate the framework's utility for scalable and structure-aware DTI prediction.

Structure-Aware Contrastive Learning with Fine-Grained Binding Representations for Drug Discovery

TL;DR

The paper tackles scalable drug–target interaction (DTI) prediction by embedding structural priors into sequence-based representations. It introduces a structure-aware protein vocabulary, SELFIES-based drug encoding, a patch-based attention trunk, and a contrastive cross-modal objective paired with a bilinear attention module to model interactions. The approach achieves state-of-the-art results on the Human and BioSNAP DTI benchmarks, remains competitive on BindingDB, and delivers superior virtual screening performance on LIT-PCBA (AUROC 68.16% and BEDROC 13.35%). Ablation and visualization analyses confirm the importance of learned aggregation, contrastive alignment, and interpretable attention over ligand–residue contacts, highlighting the method’s robustness and interpretability for scalable, structure-informed DTI pre-screening.

Abstract

Accurate identification of drug-target interactions (DTI) remains a central challenge in computational pharmacology, where sequence-based methods offer scalability. This work introduces a sequence-based drug-target interaction framework that integrates structural priors into protein representations while maintaining high-throughput screening capability. Evaluated across multiple benchmarks, the model achieves state-of-the-art performance on Human and BioSNAP datasets and remains competitive on BindingDB. In virtual screening tasks, it surpasses prior methods on LIT-PCBA, yielding substantial gains in AUROC and BEDROC. Ablation studies confirm the critical role of learned aggregation, bilinear attention, and contrastive alignment in enhancing predictive robustness. Embedding visualizations reveal improved spatial correspondence with known binding pockets and highlight interpretable attention patterns over ligand-residue contacts. These results validate the framework's utility for scalable and structure-aware DTI prediction.

Paper Structure

This paper contains 17 sections, 8 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Representations of Drugs and Protein Targets
  4. Drug and protein encoders
  5. Attention-based pooling
  6. Contrastive Learning for Drug-Target Interactions
  7. Bilinear Attention for Interaction Representation
  8. Experiments
  9. Datasets and Baselines
  10. Implementation Details
  11. Results & Discussion
  12. Performance on DTI
  13. Performance on Virtual Screening
  14. Ablation Study
  15. Visualization
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Model Architecture. Our framework encodes drug and protein sequences independently using frozen pre-trained models. The resulting embeddings are aggregated through an attention-based module and optimized via a contrastive learning objective. Cosine similarity between drug and protein representations is used to compute interaction probabilities, which support discriminative prediction of drug-target interactions and enable large-scale virtual screening.
  • Figure 2: Comparison between SaBAN and its variants on BindingDB dataset.
  • Figure 3: Visualization of feature distributions for the target CYP3A4 (a) using DrugCLIP and SaBAN models and the attention map (b) of its ligand.