ZHMolTopoRPI: A Commutative Algebra-Driven Deep Learning Framework for Robust RNA-Protein Interaction Prediction.

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Biology

Background:

• Accurate prediction of RNA-protein interactions (RPIs) is vital for understanding gene regulation.
• Current models often lack physical interpretability due to reliance on implicit embeddings.
• There is a need for interpretable computational methods for RPI prediction.

Purpose of the Study:

• To develop ZHMolTopoRPI, a computational framework integrating persistent commutative algebra and dual-tower neural networks for RPI prediction.
• To enhance the interpretability and accuracy of RPI prediction models.
• To analyze the impact of genetic variations on RPIs and their functional consequences.

Main Methods:

• Utilizing persistent Stanley-Reisner theory (PSRT) to extract multiscale mathematical features from RNA sequences.
• Employing a contrastive learning-enhanced gated attention dual-tower network (CL-GADTN) for feature fusion.
• Integrating RNA sequence features with protein semantic information from ESM2 for prediction.

Main Results:

• Achieved high MCC scores across six benchmark datasets, demonstrating robust prediction performance (e.g., 92.29% on NPInter2).
• Identified motif changes associated with pathogenic single nucleotide polymorphisms (SNPs) through commutative algebra analysis.
• Validated the framework's utility in functional target screening within the human proteome.

Conclusions:

• ZHMolTopoRPI provides a quantitative and interpretable method for precise RNA-protein interaction prediction.
• The framework enhances understanding of post-transcriptional regulation and the impact of genetic variations.
• Offers a valuable tool for functional genomics and drug discovery efforts.