CNN-SuccSite

CNN-SuccSite predicts lysine succinylation sites on proteins using a convolutional neural network and sequence-derived features to enable accurate identification of site-specific succinylation.

Key Features:

• Convolutional neural network (CNN): Implements a deep learning convolutional architecture for succinylation site prediction.
• Sequence-derived features: Utilizes position-specific amino acid composition, CKSAAP (composition of k-spaced amino acid pairs), and position-specific scoring matrices (PSSM).
• MS-verified training data: Trains on succinylated peptides identified by high-throughput mass spectrometry (MS).
• Maximal dependence decomposition (MDD): Identifies conserved substrate motifs and divides sequences into groups to enhance pattern detection.
• Evaluation by cross-validation: Assesses performance using ten-fold cross-validation.
• Independent test performance: Reports sensitivity 84.40%, specificity 86.99%, accuracy 86.79%, and Matthews correlation coefficient (MCC) 0.489 on an independent dataset (218 positives, 2621 negatives).
• Benchmarking: Demonstrates superior performance compared to existing succinylation site prediction tools.

Scientific Applications:

• Succinylation site prediction: Predicts lysine succinylation sites in protein sequences for post-translational modification mapping.
• Motif characterization: Identifies conserved substrate motifs and analyzes site-specific succinylation signatures.
• PTM-focused bioinformatics: Supports studies of post-translational modifications to inform cellular mechanism and potential therapeutic target research.

Methodology:

The method trains a convolutional neural network on MS-verified succinylated peptides using sequence-derived features (position-specific amino acid composition, CKSAAP, PSSM), applies maximal dependence decomposition (MDD) to identify and group conserved substrate motifs, and evaluates performance via ten-fold cross-validation and independent testing (218 positives, 2621 negatives).