An Ensemble Deep Learning based Predictor for Simultaneously Identifying Protein Ubiquitylation and SUMOylation...

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Protein ubiquitylation and SUMOylation are crucial post-translational modifications regulating gene expression and replication.
• Existing prediction tools often rely on feature engineering and do not address the crosstalk between these modifications.
• There is a need for integrated approaches to simultaneously predict ubiquitylation, SUMOylation, and their crosstalk sites.

Purpose of the Study:

• To develop the first all-in-one deep network for simultaneous prediction of protein Ubiquitylation and SUMOylation sites.
• To identify crosstalk sites between ubiquitylation and SUMOylation.
• To leverage deep learning for improved prediction accuracy using protein sequences and physico-chemical properties.

Main Methods:

• Developed a deep learning architecture integrating meta-classifiers and deep neural networks.
• Utilized protein sequence information and physico-chemical properties as input features.
• Trained the model in a multi-label classification mode for simultaneous prediction.

Main Results:

• Achieved promising Area Under the Curve (AUC) values: 0.838 for ubiquitylation, 0.888 for SUMOylation, and 0.862 for crosstalk sites.
• Obtained Area Precision (AP) values of 0.683, 0.804, and 0.552, respectively, validating the method's effectiveness.
• Demonstrated high performance on tenfold cross-validation.

Conclusions:

• The proposed deep network successfully classifies ubiquitylated and SUMOylated lysine residues, including crosstalk sites.
• The novel architecture outperforms existing tools for predicting ubiquitylation and SUMOylation sites.
• This integrated approach advances the study of protein post-translational modifications and their regulatory roles.