MIN: Multi-channel Interaction Network for Drug-Target Interaction with Protein Distillation

TL;DR

MIN addresses the DTI prediction problem by fusing sequence- and structure-based information through a C-Score guided representation module and a three-channel interaction network, augmented by contrastive learning to align multimodal representations. The C-Score Predictor distills residues likely to participate in binding, improving efficiency and accuracy, while the three interaction channels capture sequence-level, structure-level, and mixed-pattern interactions. On DUD-E and Human benchmarks, MIN achieves state-of-the-art performance and demonstrates explainability through residue overlap with known binding pockets and case studies. These results underscore MIN's potential to accelerate drug discovery and enhance understanding of protein binding sites in a multimodal learning framework.

Abstract

Traditional drug discovery processes are both time-consuming and require extensive professional expertise. With the accumulation of drug-target interaction (DTI) data from experimental studies, leveraging modern machine-learning techniques to discern patterns between drugs and target proteins has become increasingly feasible. In this paper, we introduce the Multi-channel Interaction Network (MIN), a novel framework designed to predict DTIs through two primary components: a representation learning module and a multi-channel interaction module. The representation learning module features a C-Score Predictor-assisted screening mechanism, which selects critical residues to enhance prediction accuracy and reduce noise. The multi-channel interaction module incorporates a structure-agnostic channel, a structure-aware channel, and an extended-mixture channel, facilitating the identification of interaction patterns at various levels for optimal complementarity. Additionally, contrastive learning is utilized to harmonize the representations of diverse data types. Our experimental evaluations on public datasets demonstrate that MIN surpasses other strong DTI prediction methods. Furthermore, the case study reveals a high overlap between the residues selected by the C-Score Predictor and those in actual binding pockets, underscoring MIN's explainability capability. These findings affirm that MIN is not only a potent tool for DTI prediction but also offers fresh insights into the prediction of protein binding sites.

Figures (6)

• Figure 1: (a) A sketch map of the drug-target interaction classification task. (b) An example illustrates that residues can be far away in the sequence but are spatially close to each other after folding. (c) The average length of proteins is much longer than that of drugs (the results are computed on the DUD-E dataset).
• Figure 2: An overview of MIN; (a) Target protein inputs; (b) Drug inputs; (c) Target protein representation learning; (d) Drug representation learning; (e) Multi-channel interaction; (f) C-Score Predictor.
• Figure 3: The conservation score (C-Score) difference between the residues in the binding pocket area and non-pocket area. A smaller C-score indicates higher conservation.
• Figure 4: Performance comparison between C-Score Predictor with other methods changed along with protein length on DUD-E, which indicates C-Score Predictor has ability to distill long proteins effectively.
• Figure 5: The comparison of the real binding pocket and prediction binding pocket. Among 37 residues in the binding pocket, 16 overlap residues between real and prediction are painted yellow.