TransformerCPI: improving compound–protein interaction prediction by sequence-based deep learning with self-attention mechanism and label reversal experiments

TLDR

Identifying compound–protein interactions is essential for drug discovery, yet many targets lack 3D structures, making sequence‑based prediction necessary, though such models suffer from dataset bias, inappropriate splits, and over‑optimistic performance estimates. The study aims to overcome these limitations by creating CPI‑specific datasets, developing the TransformerCPI neural network, and employing a rigorous label‑reversal test to verify genuine interaction learning. TransformerCPI uses a transformer architecture with self‑attention on protein sequences and compound representations, trained on the newly curated CPI datasets and evaluated through label‑reversal experiments. TransformerCPI outperforms prior methods, accurately identifies key interacting protein regions and compound atoms, and offers guidance for ligand optimization. The code and implementation are available at https://github.com/lifanchen-simm/transformerCPI.

Abstract

Abstract Motivation Identifying compound–protein interaction (CPI) is a crucial task in drug discovery and chemogenomics studies, and proteins without three-dimensional structure account for a large part of potential biological targets, which requires developing methods using only protein sequence information to predict CPI. However, sequence-based CPI models may face some specific pitfalls, including using inappropriate datasets, hidden ligand bias and splitting datasets inappropriately, resulting in overestimation of their prediction performance. Results To address these issues, we here constructed new datasets specific for CPI prediction, proposed a novel transformer neural network named TransformerCPI, and introduced a more rigorous label reversal experiment to test whether a model learns true interaction features. TransformerCPI achieved much improved performance on the new experiments, and it can be deconvolved to highlight important interacting regions of protein sequences and compound atoms, which may contribute chemical biology studies with useful guidance for further ligand structural optimization. Availability and implementation https://github.com/lifanchen-simm/transformerCPI.