Lysine lactylation as a newly discovered post-translational modification of proteins, plays a key role in various cellular processes. It can be stimulated by lactate and regulates gene expression and life activities. Mass spectrometry is currently the fundamental method for identifying post-translational modification sites, but this approach is often time-consuming and labour-intensive. In this study, we propose a hybrid deep learning model, termed TCB-Kla, for predicting human lysine lactylation sites by integrating Transformer encoder, multi-scale CNN, and bidirectional LSTM. On the independent test set, our model achieves accuracy (ACC) of 82. 1%, sensitivity (SN) of 82. 5%, specificity (SP) of 81. 7%, Matthew's correlation coefficient (MCC) of 0. 642, and area under the ROC curve (AUC) of 0. 898, with ACC, SN, AUC, and MCC exceeding those of the baseline model by 1. 8%, 3. 8%, 0. 012, and 0. 035, respectively. Additionally, we conduct 10-fold cross-validation and visualization analysis to validate the model's stability. To demonstrate the model's transfer learning capability, we select the datasets of anti-diabetic peptide for testing, and the AUC metric reaches 0. 981, surpassing the baseline model by 0. 023. Finally, to enhance the model's usability, we develop a user-friendly online web server, which can be accessed at http: //sb075813. xyz. The original datasets and codes are available at https: //github. com/ShengBin369/TCBKla.
Zhang et al. (Thu,) studied this question.