GRAM-TDI: adaptive multimodal representation learning for drug target interaction prediction

2025-09-26

AI Summary

Existing DTI prediction methods predominantly rely on SMILES–protein sequence pairs, neglecting rich multimodal biological information from both molecules and proteins. To address this, we propose MultiDTI—a unified four-modal representation framework integrating molecular graphs, SMILES strings, protein sequences, and 3D structural features. MultiDTI introduces high-order contrastive learning for cross-modal semantic alignment and an adaptive modality dropout mechanism for dynamic weight modulation. It is the first to incorporate IC50-based weak supervision to enhance biological interpretability, and synergistically combines pretrained encoders with a dedicated multimodal fusion network to improve generalization. Evaluated on four benchmark datasets—BindingDB, DrugBank, DAVIS, and KIBA—MultiDTI achieves significant improvements over state-of-the-art methods in both accuracy and robustness, with verified cross-dataset transferability.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Drug target interaction (DTI) prediction is a cornerstone of computational drug discovery, enabling rational design, repurposing, and mechanistic insights. While deep learning has advanced DTI modeling, existing approaches primarily rely on SMILES protein pairs and fail to exploit the rich multimodal information available for small molecules and proteins. We introduce GRAMDTI, a pretraining framework that integrates multimodal molecular and protein inputs into unified representations. GRAMDTI extends volume based contrastive learning to four modalities, capturing higher-order semantic alignment beyond conventional pairwise approaches. To handle modality informativeness, we propose adaptive modality dropout, dynamically regulating each modality's contribution during pre-training. Additionally, IC50 activity measurements, when available, are incorporated as weak supervision to ground representations in biologically meaningful interaction strengths. Experiments on four publicly available datasets demonstrate that GRAMDTI consistently outperforms state of the art baselines. Our results highlight the benefits of higher order multimodal alignment, adaptive modality utilization, and auxiliary supervision for robust and generalizable DTI prediction.

Problem

• Predicts drug-target interactions using multimodal molecular representations
• Integrates four molecular modalities with adaptive contribution control
• Improves prediction accuracy through contrastive learning and activity supervision

Innovation

• Integrates multimodal molecular and protein inputs
• Extends contrastive learning to four modalities
• Uses adaptive modality dropout during pre-training