CLW_SUMO: A hybrid deep learning model for predicting protein SUMOylation sites

Abstract

Protein SUMOylation is one of the most important post-translational modifications in Eukaryotes species and plays significant roles in many biological processes. The mechanism underlined the SUMOylation process will be an important cause leading to many common serious diseases, such as breast cancer, cardiac, Parkinson’s, Alzheimer’s disease, etc. Due to the very important roles regulated by SUMOylation, the demand for an in-depth understanding of SUMOylation and its mechanism is currently a hot topic that interests many scientists. In this study, we propose a novel approach, called CLW-SUMO, for predicting SUMOylation sites using a hybrid deep learning model that combines convolutional neural networks (CNN) and long short-term memory (LSTM), using Word2Vec as the word embedding technique. The 10-fold cross-validation demonstrates that our proposed model achieves the best performance with an accuracy of 82.33%, MCC of 0.589 and AUC of 0.829. Besides, the independent testing also shows that our proposed model obtains the highest performance, reaching an accuracy of 90.03%, MCC of 0.773 and AUC of 0.889. Furthermore, when compared to several existing predictors of SUMOylation using an independent dataset, our proposed model exhibits the highest performance with an ACC value of 90.03% and an MCC value of 0.773. We hope that our findings will provide effective suggestions and greatly help researchers in their studies related to protein SUMOylation identification.