Prediction of S-Palmitoylation Sites in the Male/Female Mouse Using the Protein Language Model

S-palmitoylation, i.e. post-translational modification of cysteine thiol side chain, is crucial in various biological processes and human diseases. Therefore, identification of S-palmitoylation sites from protein sequences is important, especially for understanding their functional consequences. Deep language models have shown impressive results in natural language processing tasks. Recently, they have also been used to biological sequences. Major objective of this paper is to examine whether deep language models can identify S-palmitoylation cites more efficiently. Three categories of synaptic protein datasets were considered for this experiment: male mouse, female mouse, and combination of both. Weighted data samples from each group was used for training, while held-out data was used for testing and performance comparison purpose. The proposed method performed much better than the state-of-the-art approaches. Accuracy improvements on the hold-out dataset are male—5%, female—8%, and combined—14%. One-star consensus strategy has been used for final classification where performance improved significantly (more than 20% for all three types).