Sul-BertGRU: an ensemble deep learning method integrating information entropy-enhanced BERT and directional multi-GRU for S-sulfhydration sites prediction

• DOI: 10.1093/bioinformatics/btaf078
• Publication Date: 2025-02-20T18:07:41Z
• Authors: Xirun Wei; Qiao Ning; Kuiyang Che; Zhaowei Liu; Hui Li; Shikai Guo
• Keywords: Ensemble Learning

S-sulfhydration, a crucial post-translational protein modification, is pivotal in cellular recognition, signaling processes, and the development progression of cardiovascular neurological disorders, so identifying S-sulfhydration sites for studies cell biology. Deep learning shows high efficiency accuracy compared to traditional methods that often lack sensitivity specificity accurately locating nonsulfhydration sites. Therefore, we employ deep tackle challenge pinpointing In this work, introduce approach called Sul-BertGRU, designed specifically predicting proteins, integrates multi-directional gated recurrent unit (GRU) BERT. First, Sul-BertGRU proposes an information entropy-enhanced BERT (IE-BERT) preprocess sequences extract initial features. Subsequently, confidence employed eliminate potential samples from select reliable negative samples. Then, considering directional nature modification process, are categorized into left, right, full centred on cysteines. We build GRU enhance extraction sequence features model details enzymatic reaction involved S-sulfhydration. Ultimately, apply parallel multi-head self-attention mechanism alongside convolutional neural network (CNN) deeply analyze might be missed at local level. achieves sensitivity, specificity, precision, accuracy, Matthews correlation coefficient, area under curve scores 85.82%, 68.24%, 74.80%, 77.44%, 55.13%, 77.03%, respectively. demonstrates exceptional performance proves method The source code data available https://github.com/Severus0902/Sul-BertGRU/. Supplementary Bioinformatics online.