LMPhosSite: A Deep Learning-Based Approach for General Protein Phosphorylation Site Prediction Using Embeddings from the Local Window Sequence and Pretrained Protein Language Model

Phosphorylation is one of the most important post-translationalmodifications and plays a pivotal role in various cellular processes.Although there exist several computational tools to predict phosphorylationsites, existing tools have not yet harnessed the knowledge distilledby pretrained protein language models. Herein, we present a noveldeep learning-based approach called LMPhosSite for the general phosphorylationsite prediction that integrates embeddings from the local window sequenceand the contextualized embedding obtained using global (overall) proteinsequence from a pretrained protein language model to improve the predictionperformance. Thus, the LMPhosSite consists of two base-models: onefor capturing effective local representation and the other for capturingglobal per-residue contextualized embedding from a pretrained proteinlanguage model. The output of these base-models is integrated usinga score-level fusion approach. LMPhosSite achieves a precision, recall,Matthew's correlation coefficient, and F1-score of 38.78%, 67.12%,0.390, and 49.15%, for the combined serine and threonine independenttest data set and 34.90%, 62.03%, 0.298, and 44.67%, respectively,for the tyrosine independent test data set, which is better than thecompared approaches. These results demonstrate that LMPhosSite isa robust computational tool for the prediction of the general phosphorylationsites in proteins.