Characterization and Identification of Lysine Succinylation Sites based on Deep Learning Method

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Succinylation is a crucial protein post-translational modification (PTM) involved in cellular processes.
• Existing computational tools for succinylation site prediction primarily rely on traditional machine learning methods.
• There is a need for advanced prediction models leveraging deep learning for improved accuracy.

Purpose of the Study:

• To develop a deep learning-based model for accurate prediction of protein succinylation sites.
• To investigate substrate site specificity using sequence-based attributes and motif discovery.
• To compare the performance of the deep learning model against existing prediction tools.

Main Methods:

• Utilized sequence-based attributes including position-specific amino acid composition, CKSAAP, and PSSM.
• Employed Maximal Dependence Decomposition (MDD) for substrate motif detection.
• Developed and validated a deep learning model, tested using ten-fold cross-validation and an independent dataset.

Main Results:

• The deep learning model trained with PSSM and CKSAAP attributes achieved superior predictive performance.
• The proposed model demonstrated better accuracy and efficiency compared to traditional machine learning methods.
• Independent testing showed high sensitivity (84.40%), specificity (86.99%), and accuracy (86.79%) with an MCC of 0.489.

Conclusions:

• The developed deep learning model offers a significant advancement in predicting succinylation sites.
• The CNN-SuccSite web tool provides a valuable, freely accessible resource for researchers.
• This work enhances the understanding of succinylation site specificity and computational prediction capabilities.