Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

M3ST-DTI: A multi-task learning model for drug-target interactions based on multi-modal features and multi-stage alignment

Xiangyu Li, Ran Su, Liangliang Liu

TL;DR

DTI prediction is hindered by weak cross-modal alignment and limited intra-modal interactions. M3ST-DTI tackles this by constructing textual, structural, and functional features and aligning them through Gram Loss and MCA in an early fusion stage, followed by deep intra-modal fusion with BCA, HGCNF, and a DOF module, under a multi-task objective. On BindingDB, it achieves state-of-the-art performance across Accuracy, F1, AUROC, and AUPRC, with ablations confirming the necessity of both early and late fusion and cross-modal interaction components; UMAP visualizations corroborate improved class separability during fusion. By integrating diverse modalities in a staged, interpretable framework, M3ST-DTI enhances generalization and offers a practical pathway for improving drug discovery and precision therapeutics.

Abstract

Accurate prediction of drug-target interactions (DTI) is pivotal in drug discovery. However, existing approaches often fail to capture deep intra-modal feature interactions or achieve effective cross-modal alignment, limiting predictive performance and generalization. To address these challenges, we propose M3ST-DTI, a multi-task learning model that enables multi-stage integration and alignment of multi modal features for DTI prediction. M3ST-DTI incorporates three types of features-textual, structural, and functional and enhances intra-modal representations using self-attention mechanisms and a hybrid pooling graph attention module. For early-stage feature alignment and fusion, the model in tegrates MCA with Gram loss as a structural constraint. In the later stage, a BCA module captures fine-grained interactions between drugs and targets within each modality, while a deep orthogonal fusion module mitigates feature redundancy.Extensive evaluations on benchmark datasets demonstrate that M3ST-DTI consistently outperforms state-of-the art methods across diverse metrics

M3ST-DTI: A multi-task learning model for drug-target interactions based on multi-modal features and multi-stage alignment

TL;DR

DTI prediction is hindered by weak cross-modal alignment and limited intra-modal interactions. M3ST-DTI tackles this by constructing textual, structural, and functional features and aligning them through Gram Loss and MCA in an early fusion stage, followed by deep intra-modal fusion with BCA, HGCNF, and a DOF module, under a multi-task objective. On BindingDB, it achieves state-of-the-art performance across Accuracy, F1, AUROC, and AUPRC, with ablations confirming the necessity of both early and late fusion and cross-modal interaction components; UMAP visualizations corroborate improved class separability during fusion. By integrating diverse modalities in a staged, interpretable framework, M3ST-DTI enhances generalization and offers a practical pathway for improving drug discovery and precision therapeutics.

Abstract

Accurate prediction of drug-target interactions (DTI) is pivotal in drug discovery. However, existing approaches often fail to capture deep intra-modal feature interactions or achieve effective cross-modal alignment, limiting predictive performance and generalization. To address these challenges, we propose M3ST-DTI, a multi-task learning model that enables multi-stage integration and alignment of multi modal features for DTI prediction. M3ST-DTI incorporates three types of features-textual, structural, and functional and enhances intra-modal representations using self-attention mechanisms and a hybrid pooling graph attention module. For early-stage feature alignment and fusion, the model in tegrates MCA with Gram loss as a structural constraint. In the later stage, a BCA module captures fine-grained interactions between drugs and targets within each modality, while a deep orthogonal fusion module mitigates feature redundancy.Extensive evaluations on benchmark datasets demonstrate that M3ST-DTI consistently outperforms state-of-the art methods across diverse metrics
Paper Structure (18 sections, 8 equations, 4 figures, 1 table)

This paper contains 18 sections, 8 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Benchmark Datasets
  4. Multi-modal Feature Construction
  5. Multi-modal feature extraction stage
  6. Text Feature Extraction
  7. Structure Feature Extraction
  8. Function Feature Extraction
  9. Multi-level Feature Fusion
  10. Early Fusion Stage
  11. Late Fusion Stage
  12. Interaction Prediction
  13. Results
  14. Experimental Settings
  15. Ablation Experiment
  16. ...and 3 more sections

Figures (4)

  • Figure 1: The process of constructing multi-modal features from a standard DTI dataset.(a) the process of constructing textual features (b) the process of constructing structural features (c) the process of constructing functional featuresn.
  • Figure 2: Architecture flow of $M^3$ST-DTI method: (a) multi-modal feature extraction stage (b) early fusion stage (c) late fusion stage (d) interaction prediction stage.
  • Figure 3: Radial Plot of the Comparison Experiment Results
  • Figure 4: UMAP Visualization of feature distributions