ResUbiNet: A Novel Deep Learning Architecture for Ubiquitination Site Prediction.

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Genomics

Background:

• Ubiquitination is a critical post-translational modification regulating key cellular processes like protein degradation, signal transduction, DNA repair, and cell cycle control.
• Accurate identification of ubiquitination sites is essential for elucidating ubiquitination mechanisms and understanding diseases linked to ubiquitination dysregulation.

Purpose of the Study:

• To develop a novel deep learning architecture for accurate prediction of ubiquitination sites.
• To enhance the understanding of ubiquitination mechanisms and associated disease pathogenesis.

Main Methods:

• Developed ResUbiNet, a deep learning model integrating a protein language model (ProtTrans), amino acid properties, and the BLOSUM62 matrix for sequence embedding.
• Employed advanced architectural components including transformer, multi-kernel convolution, residual connection, and squeeze-and-excitation for robust feature extraction.

Main Results:

• ResUbiNet demonstrated superior prediction performance.
• Achieved better results compared to established models like hCKSAAP_UbSite, RUBI, MDCapsUbi, and MusiteDeep in cross-validation and external tests.

Conclusions:

• ResUbiNet significantly improves ubiquitination site prediction accuracy.
• The model's advanced features and architecture facilitate a deeper understanding of ubiquitination and its role in disease.