Search

Lysine post-translational modification (PTM) is involved in almost all cellular activities, and plays a critical role in protein structure and function. In the past decades, a variety of experimental methods for identifying PTM sites have emerged, which has promoted the development of PTM research. However, the experimental identification method is time-consuming and labor-intensive. Although deep learning has shown great potential in PTM site prediction recently, existing deep learning-based methods have limited effectiveness in predicting species-specific PTM sites. In this article, we propose a semi-supervised domain adaptation method, named Semi-ssPTM, for species-specific lysine PTM site prediction. Semi-ssPTM improves performance by introducing unlabeled PTM data and effectively enhances the generalization ability of species-specific prediction models. In addition, we train and test predictive models on two common lysine PTM acetylation and ubiquitination datasets. The experimental results show that our proposed semi-supervised method can effectively improve the prediction performance of multiple species and is superior to existing lysine PTM site prediction tools. The web server of Semi-ssPTM, as well as all datasets and associated code used in this study, is freely available at (http://bmi.ahu.edu.cn/Semi-ssPTM/). We expect Semi-ssPTM will provide practical guidance and useful insights for PTM site predictions and inspire future bioinformatics research in related fields.