Promoters are key DNA elements that regulate bacterial gene expression, yet most existing computational methods demonstrate limited effectiveness in predicting promoters across diverse bacterial species. Here, we propose PBPICBA, a deep learning model featuring a dual-path architecture that integrates two-dimensional convolution and improved Convolutional Block Attention Module for accurate species-specific bacterial promoter identification. The model employs a comprehensive encoding scheme combining one-hot encoding, Nucleotide Chemical Property C2, and ESM-2 representations. Evaluation on 13 species-specific bacterial promoter datasets shows that PBPICBA achieves superior performance in 11 species. This study provides a robust framework for species-specific bacterial promoter prediction and enhances our understanding of transcriptional regulatory mechanisms. Research data is available in this public repository: https: //github. com/liuchang-chun /PBPICBAA.
Wang et al. (Thu,) studied this question.