T4SEpp: a pipeline integrated with protein language models effectively predicting bacterial type IV secreted effectors

Abstract

AbstractMany pathogenic bacteria use type IV secretion systems(T4SSs) to deliver effectors (T4SEs) into the cytoplasm of eukaryotic cells, causeing diseases. The identification of effectors is a crucial step in understanding the mechanisms of bacterial pathogenicity, but this remains a major challenge. In this study, we used the full-length embedding features generated by six pre-trained protein language models to train classifiers predicting T4SEs, and compared their performance. An integrated model T4SEpp was assembled by a module searching full-length, signal sequence and effector domain homologs of known T4SEs, a machine learning module based on the hand-crafted features extracted from the signal sequences, and the third module containing three best-performing protein language pre-trained models. T4SEpp outperformed the other state-of-the-art (SOTA) software tools, achieving ∼0.95 sensitivity at a high specificity of ∼0.99, based on the assessment of an independent testing dataset. Additionally, we performed a comprehensive search among 8,761 bacterial species, leading to the discovery of 227 species belonging to 3 phyla and 117 genera that possess T4SSs. Furthermore, leveraging the power of T4SEpp, we successfully identified a grand total of 12,622 plausible T4SEs. Overall, T4SEpp provides a better solution to assist in the identification of bacterial T4SEs, and facilitates studies of bacterial pathogenicity. T4SEpp is freely accessible athttps://bis.zju.edu.cn/T4SEpp.