Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SeekRBP: Leveraging Sequence-Structure Integration with Reinforcement Learning for Receptor-Binding Protein Identification

Xiling Luo, Le Ou-Yang, Yang Shen, Jiaojiao Guan, Dehan Cai, Jun Zhang, Rui Zhang, Yanni Sun, Jiayu Shang

TL;DR

SeekRBP is presented, a sequence--structure framework that models negative sampling as a sequential decision-making problem and dynamically prioritizes informative non-RBP sequences based on real-time training feedback, complemented by a multimodal fusion of protein language and structural embeddings.

Abstract

Motivation: Receptor-binding proteins (RBPs) initiate viral infection and determine host specificity, serving as key targets for phage engineering and therapy. However, the identification of RBPs is complicated by their extreme sequence divergence, which often renders traditional homology-based alignment methods ineffective. While machine learning offers a promising alternative, such approaches struggle with severe class imbalance and the difficulty of selecting informative negative samples from heterogeneous tail proteins. Existing methods often fail to balance learning from these ``hard negatives'' while maintaining generalization. Results: We present SeekRBP, a sequence--structure framework that models negative sampling as a sequential decision-making problem. By employing a multi-armed bandit strategy, SeekRBP dynamically prioritizes informative non-RBP sequences based on real-time training feedback, complemented by a multimodal fusion of protein language and structural embeddings. Benchmarking demonstrates that SeekRBP consistently outperforms static sampling strategies. Furthermore, a case study on Vibrio phages validates that SeekRBP effectively identifies RBPs to improve host prediction, highlighting its potential for large-scale annotation and synthetic biology applications.

SeekRBP: Leveraging Sequence-Structure Integration with Reinforcement Learning for Receptor-Binding Protein Identification

TL;DR

SeekRBP is presented, a sequence--structure framework that models negative sampling as a sequential decision-making problem and dynamically prioritizes informative non-RBP sequences based on real-time training feedback, complemented by a multimodal fusion of protein language and structural embeddings.

Abstract

Motivation: Receptor-binding proteins (RBPs) initiate viral infection and determine host specificity, serving as key targets for phage engineering and therapy. However, the identification of RBPs is complicated by their extreme sequence divergence, which often renders traditional homology-based alignment methods ineffective. While machine learning offers a promising alternative, such approaches struggle with severe class imbalance and the difficulty of selecting informative negative samples from heterogeneous tail proteins. Existing methods often fail to balance learning from these ``hard negatives'' while maintaining generalization. Results: We present SeekRBP, a sequence--structure framework that models negative sampling as a sequential decision-making problem. By employing a multi-armed bandit strategy, SeekRBP dynamically prioritizes informative non-RBP sequences based on real-time training feedback, complemented by a multimodal fusion of protein language and structural embeddings. Benchmarking demonstrates that SeekRBP consistently outperforms static sampling strategies. Furthermore, a case study on Vibrio phages validates that SeekRBP effectively identifies RBPs to improve host prediction, highlighting its potential for large-scale annotation and synthetic biology applications.
Paper Structure (25 sections, 9 equations, 5 figures)

This paper contains 25 sections, 9 equations, 5 figures.

Table of Contents

  1. Introduction
  2. Existing work
  3. Overview
  4. Materials and methods
  5. Bandit-based adaptive negative sampling framework
  6. Dataset construction
  7. Bandit-based negative sample selection and utility estimation
  8. Model architecture
  9. Dual-branch protein encoding
  10. The 1D sequence branch
  11. The 3D structure branch
  12. Adaptive expert fusion module (AEFM)
  13. Additive Interaction Path:
  14. Multiplicative Interaction Path:
  15. Channel-wise Adaptive Mixing:
  16. ...and 10 more sections

Figures (5)

  • Figure 1: Overview of the Multi-Armed Bandit (MAB)-driven dynamic negative sampling pipeline. A UCB1-based multi-armed bandit strategy iteratively selects informative hard negatives from a large pool and updates their utility using EL2N rewards after each training round. Sequence and structure representations are adaptively fused and fed into a classifier to predict RBP versus non-RBP.
  • Figure 2: Pipeline of the proposed model for phage RBP identification. The input protein is encoded by a sequence branch and a structure branch, whose features are integrated by an Adaptive Expert Fusion Module combining gated additive and low-rank bilinear interactions. The fused representation is fed into a classifier to predict RBP versus non-RBP.
  • Figure 3: ROC curve comparison of ablation study. A) sampling strategies. The w/o Exploration variant represents a degenerate form of the bandit-based strategy. B) different feature modalities. Sequence only uses sequence-derived features, structure only uses structure-derived features, and sequence combined with structure integrates both modalities. C) feature fusion strategies. Sum: additive fusion; Concat: concatenation; AEFM: adaptive expert fusion module.
  • Figure 4: Comparison with benchmark methods on phage RBP identification. Precision, Recall, and F1-score are reported for PhANNs, PhageRBPdetection, Blastp, Pharokka, and our method, showing overall improved performance of the proposed approach.
  • Figure 5: TM-score comparison between predicted RBPs (Predicted) and manually selected RBPs (Observed) in the Vibrio dataset. The predicted RBPs show larger TM-scores range, possibly reflecting discovered divergent RBP families.