HyboWaveNet: Hyperbolic Graph Neural Networks with Multi-Scale Wavelet Transform for Protein-Protein Interaction Prediction

TL;DR

HyboWaveNet tackles PPI prediction by embedding proteins in hyperbolic Lorentz space to naturally capture hierarchical structures, and by applying a multiscale graph wavelet transform to extract local-to-global interaction patterns. The model combines a Lorentz-space GNN encoder with random-walk based wavelet features and a contrastive learning objective, then scores potential interactions using a Lorentz-space distance. Empirical results on a Luo/HPRD-derived dataset show HyboWaveNet outperforms several baselines, with ablations confirming the contributions of the hyperbolic encoding, multiscale wavelets, and contrastive learning. This work links geometric deep learning and signal processing to enhance interpretability and generalization in protein interaction prediction.

Abstract

Protein-protein interactions (PPIs) are fundamental for deciphering cellular functions,disease pathways,and drug discovery.Although existing neural networks and machine learning methods have achieved high accuracy in PPI prediction,their black-box nature leads to a lack of causal interpretation of the prediction results and difficulty in capturing hierarchical geometries and multi-scale dynamic interaction patterns among proteins.To address these challenges, we propose HyboWaveNet,a novel deep learning framework that collaborates with hyperbolic graphical neural networks (HGNNs) and multiscale graphical wavelet transform for robust PPI prediction. Mapping protein features to Lorentz space simulates hierarchical topological relationships among biomolecules via a hyperbolic distance metric,enabling node feature representations that better fit biological a priori.HyboWaveNet inherently simulates hierarchical and scale-free biological relationships, while the integration of wavelet transforms enables adaptive extraction of local and global interaction features across different resolutions. Our framework generates node feature representations via a graph neural network under the Lorenz model and generates pairs of positive samples under multiple different views for comparative learning, followed by further feature extraction via multi-scale graph wavelet transforms to predict potential PPIs. Experiments on public datasets show that HyboWaveNet improves over both existing state-of-the-art methods. We also demonstrate through ablation experimental studies that the multi-scale graph wavelet transform module improves the predictive performance and generalization ability of HyboWaveNet. This work links geometric deep learning and signal processing to advance PPI prediction, providing a principled approach for analyzing complex biological systems

Figures (5)

• Figure 1: HyboWaveNet workflow. (a) Project the feature matrix into the tangent space, and then project from the tangent space into the hyperbolic space to generate the feature matrix under the hyperbolic space, and input it into the multilayer GCN to do comparison learning by using the attention mechanism to aggregate the features extracted from each layer of GCN. (b) Calculate the random wandering matrix P by the adjacency matrix, then derive the wavelet coefficients from P and the scale list, and finally multiply them with the feature matrix aggregated by the attention to get the final features to obtain the PPI prediction.
• Figure 2: Values of the two assessment metrics, AUC and AP, with different encoders
• Figure 3: Values of AUC and AP for two evaluations with different encoders with and without multiscale wavelet transform and comparison learning module
• Figure 4: Values of AUC and AP under different scale lists
• Figure 5: The 10 features in the initialized node features that contribute the most to PPI prediction