DeepSSPred: A Deep Learning Based Sulfenylation Site Predictor Via a Novel nSegmented Optimize Federated Feature...

Area of Science:

• Bioinformatics
• Computational Biology
• Post-Translational Modifications

Background:

• S-sulfenylation is a crucial post-translational modification involved in various cellular processes.
• Existing computational models for predicting S-sulfenylation sites have limitations in performance due to feature encoding and data imbalance.

Purpose of the Study:

• To develop a novel and robust computational predictor for accurately discriminating S-sulfenylation (SC) sites from non-SC sites.

Main Methods:

• Developed DeepSSPred, an innovative bioinformatics tool utilizing nSegmented hybrid features.
• Employed Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance.
• Utilized a 2D-Convolutional Neural Network (2D-CNN) classifier with 10-fold jackknife cross-validation.

Main Results:

• DeepSSPred demonstrated superior performance compared to existing methods on both training and independent datasets.
• Achieved significant improvements in accuracy, sensitivity (Sn), specificity (Sp), and Matthews Correlation Coefficient (MCC).
• The model's performance was validated through rigorous cross-validation, showing enhanced predictive power.

Conclusions:

• DeepSSPred is an effective sequence-based automated predictor for S-sulfenylation sites.
• The predictor's efficacy stems from novel feature encoding, SMOTE, and a tuned 2D-CNN classifier.
• This tool offers valuable insights for understanding S-sulfenylation and aids in pharmaceutical drug design.