Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Extracting Inter-Protein Interactions Via Multitasking Graph Structure Learning

Jiang Li, Yuan-Ting Li

TL;DR

The paper tackles protein-protein interaction (PPI) prediction by leveraging internal protein structure through graph neural networks. It proposes MgslaPPI, a two-stage multitask graph learning framework comprising amino acid residue reconstruction (A2RR) with a graph attention network and protein interaction prediction (PIP) with a GCNII encoder, augmented by two auxiliary tasks: protein feature reconstruction (PFR) and masked interaction prediction (MIP). Empirical results on SHS27K and SHS148K under BFS, DFS, and Random partitions show MgslaPPI outperforms state-of-the-art baselines, with ablations confirming the contribution of each component. The approach reduces memory overhead while improving the encoder’s expressiveness, yielding robust performance on unseen proteins and offering practical benefits for structure-aware PPI prediction.

Abstract

Identifying protein-protein interactions (PPI) is crucial for gaining in-depth insights into numerous biological processes within cells and holds significant guiding value in areas such as drug development and disease treatment. Currently, most PPI prediction methods focus primarily on the study of protein sequences, neglecting the critical role of the internal structure of proteins. This paper proposes a novel PPI prediction method named MgslaPPI, which utilizes graph attention to mine protein structural information and enhances the expressive power of the protein encoder through multitask learning strategy. Specifically, we decompose the end-to-end PPI prediction process into two stages: amino acid residue reconstruction (A2RR) and protein interaction prediction (PIP). In the A2RR stage, we employ a graph attention-based residue reconstruction method to explore the internal relationships and features of proteins. In the PIP stage, in addition to the basic interaction prediction task, we introduce two auxiliary tasks, i.e., protein feature reconstruction (PFR) and masked interaction prediction (MIP). The PFR task aims to reconstruct the representation of proteins in the PIP stage, while the MIP task uses partially masked protein features for PPI prediction, with both working in concert to prompt MgslaPPI to capture more useful information. Experimental results demonstrate that MgslaPPI significantly outperforms existing state-of-the-art methods under various data partitioning schemes.

Extracting Inter-Protein Interactions Via Multitasking Graph Structure Learning

TL;DR

The paper tackles protein-protein interaction (PPI) prediction by leveraging internal protein structure through graph neural networks. It proposes MgslaPPI, a two-stage multitask graph learning framework comprising amino acid residue reconstruction (A2RR) with a graph attention network and protein interaction prediction (PIP) with a GCNII encoder, augmented by two auxiliary tasks: protein feature reconstruction (PFR) and masked interaction prediction (MIP). Empirical results on SHS27K and SHS148K under BFS, DFS, and Random partitions show MgslaPPI outperforms state-of-the-art baselines, with ablations confirming the contribution of each component. The approach reduces memory overhead while improving the encoder’s expressiveness, yielding robust performance on unseen proteins and offering practical benefits for structure-aware PPI prediction.

Abstract

Identifying protein-protein interactions (PPI) is crucial for gaining in-depth insights into numerous biological processes within cells and holds significant guiding value in areas such as drug development and disease treatment. Currently, most PPI prediction methods focus primarily on the study of protein sequences, neglecting the critical role of the internal structure of proteins. This paper proposes a novel PPI prediction method named MgslaPPI, which utilizes graph attention to mine protein structural information and enhances the expressive power of the protein encoder through multitask learning strategy. Specifically, we decompose the end-to-end PPI prediction process into two stages: amino acid residue reconstruction (A2RR) and protein interaction prediction (PIP). In the A2RR stage, we employ a graph attention-based residue reconstruction method to explore the internal relationships and features of proteins. In the PIP stage, in addition to the basic interaction prediction task, we introduce two auxiliary tasks, i.e., protein feature reconstruction (PFR) and masked interaction prediction (MIP). The PFR task aims to reconstruct the representation of proteins in the PIP stage, while the MIP task uses partially masked protein features for PPI prediction, with both working in concert to prompt MgslaPPI to capture more useful information. Experimental results demonstrate that MgslaPPI significantly outperforms existing state-of-the-art methods under various data partitioning schemes.

Paper Structure

This paper contains 18 sections, 15 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Task definition
  5. Amino acid residue reconstruction
  6. Protein interaction prediction
  7. Auxiliary task
  8. Experiment
  9. Dataset
  10. Baseline
  11. Implementation detail
  12. Main result
  13. Ablation study
  14. In-depth analysis of different subsets
  15. Result for each category
  16. ...and 3 more sections

Figures (5)

  • Figure 1: Overall workflow of the proposed MgslaPPI.
  • Figure 2: Results on different subsets (BS, ES, and NS) of the SHS27k dataset. F1 scores of MAPE-PPI are reproduced based on the official code wu2024mapeppi, and W-Ave denotes the weighted average.
  • Figure 3: Accuracy for each category on the SHS27k dataset.
  • Figure 4: Performance variation with network depth on the SHS27k dataset.
  • Figure 5: Performance variation with network depth on the SHS27k dataset.