PSC-CPI: Multi-Scale Protein Sequence-Structure Contrasting for Efficient and Generalizable Compound-Protein Interaction Prediction

TL;DR

PSC-CPI tackles the compound–protein interaction prediction problem under modality missing and domain shift by introducing a multi-scale sequence–structure contrasting framework. It pre-trains sequence and structure encoders with intra- and cross-modality contrastive objectives and applies length-variable augmentation to capture multi-scale information. The approach yields strong generalization across four data splits, especially the challenging Unseen-Both setting, and remains effective when inferring from a single protein modality. Empirically, PSC-CPI achieves enhanced pattern and strength predictions over baselines and demonstrates robustness to protein length and atom count, supporting scalable CPI screening.

Abstract

Compound-Protein Interaction (CPI) prediction aims to predict the pattern and strength of compound-protein interactions for rational drug discovery. Existing deep learning-based methods utilize only the single modality of protein sequences or structures and lack the co-modeling of the joint distribution of the two modalities, which may lead to significant performance drops in complex real-world scenarios due to various factors, e.g., modality missing and domain shifting. More importantly, these methods only model protein sequences and structures at a single fixed scale, neglecting more fine-grained multi-scale information, such as those embedded in key protein fragments. In this paper, we propose a novel multi-scale Protein Sequence-structure Contrasting framework for CPI prediction (PSC-CPI), which captures the dependencies between protein sequences and structures through both intra-modality and cross-modality contrasting. We further apply length-variable protein augmentation to allow contrasting to be performed at different scales, from the amino acid level to the sequence level. Finally, in order to more fairly evaluate the model generalizability, we split the test data into four settings based on whether compounds and proteins have been observed during the training stage. Extensive experiments have shown that PSC-CPI generalizes well in all four settings, particularly in the more challenging ``Unseen-Both" setting, where neither compounds nor proteins have been observed during training. Furthermore, even when encountering a situation of modality missing, i.e., inference with only single-modality protein data, PSC-CPI still exhibits comparable or even better performance than previous approaches.

Figures (7)

• Figure 1: An illustration of Compound-Protein Interaction.
• Figure 2: A high-level illustration of multi-scale protein sequence-structure contrasting framework for CPI prediction.
• Figure 3: Illustration of multi-scale protein sequence-structure contrastive framework, where a length-variable augmentation module is used to generate subsequences $\{\mathcal{S}^{(i,k)}\}_{k=1}^K$ of different lengths and corresponding subgraphs $\{\mathcal{G}_P^{(i,k)}\}_{k=1}^K$, which are then encoded separately by sequence and structure encoders to perform intra- and cross-modality contrasting at different scales.
• Figure 4: (a) Pre-training on known protein sequence-structure pairs by multi-scale contrasting. (b)(c)(d) Three inference settings where only protein sequences, structures, or both modalities are provided.
• Figure 5: Histogram of proteins and compounds on Karimi.