Enhancing Protein-Protein Interaction Prediction with Hierarchical Motif-based Multimodal Protein Embedding

Protein-protein interactions (PPIs) are essential for many biological processes. However, existing PPI prediction approaches suffer from two major limitations: they overlook the hierarchical organization of proteins, particularly meso-scale motifs that critically regulate PPIs, and fail to effectively integrate sequence, structure, and function modalities. To address these limitations, we propose MMM-PPI, a Hierarchical Motif-based Multi-Modal protein Encoder for PPI Prediction that constructs PPI embeddings in a bottom-up multi-modal manner across three scales. At the micro-scale, we encode three modal residue features; at the meso-scale, a novel multimodal motif encoder aggregates residues into spatially-informed motif embeddings; at the macro-scale, a multimodal protein encoder integrates motifs into protein embeddings by jointly modeling motif importance and inter-modal correlations. The pre-trained encoder can be used off-the-shelf for large-scale PPI prediction. Extensive experiments on multiple PPI datasets show that MMM-PPI outperforms state-of-the-art multi-label PPI prediction models, particularly under challenging data partitions and limited data scenarios. Codes are in https://github.com/yzf-code/MMM-PPI.

Proteins are the molecular machines that carry out almost every task in living cells, and they rarely work alone — they interact with other proteins to perform their functions. Being able to predict which proteins interact with each other (and how) is therefore extremely important for understanding biology and developing new drugs. However, current computer-based prediction methods have two main weaknesses. First, they tend to treat a protein either as a long string of letters (its sequence) or as a single 3D shape, ignoring the fact that proteins are organized in layers — from individual building blocks (residues), to small functional patterns (motifs), to the whole protein. These mid-sized patterns, called motifs, are especially important because they often determine where and how two proteins stick together. Second, existing methods usually focus on just one type of information about a protein, even though sequence, 3D structure, and biological function each tell us something different and useful. To overcome these problems, we developed MMM-PPI, a new method that builds up its understanding of a protein step by step, from small to large, while combining three kinds of information at every step. It first looks at each residue using sequence, structure, and function clues, then groups nearby residues into motifs, and finally combines the motifs into a complete picture of the protein — paying attention to which motifs matter most and how the different types of information relate to each other. Once trained, our model can be applied directly to predict protein interactions on a large scale. In tests on several standard datasets, MMM-PPI consistently outperforms the best existing methods, especially in difficult situations such as when training data is limited or when the test proteins are very different from those seen during training.