NetCGlyc

NetCGlyc predicts C-mannosylation sites in mammalian proteins, identifying tryptophan residues modified by an alpha-mannopyranose via a carbon–carbon linkage and characterizing sequence and structural determinants of this post-translational modification.

Key Features:

• Neural Network Approach: Employs a neural network model trained on sequence data to predict C-mannosylation sites.
• Sequence Context Analysis: Uses a 21-residue sequence input window and incorporates the presence or absence of the WXXW motif while detecting preferences such as acceptance of cysteine at position +3 and small/polar residues (Ser, Ala, Gly, Thr) at position +1; only ~two-thirds of known sites follow the WXXW motif.
• Structural Insights: Integrates Protein Data Bank structural data showing that modified tryptophan residues are often partially solvent-exposed.
• Predictive Performance: Reports 93% accuracy for both positive (mannosylated) and negative (non-mannosylated) sites and indicates the consensus WXXW motif alone identifies 67% of positive sites.

Scientific Applications:

• Protein modification studies: Enables identification of candidate C-mannosylation sites to study functional effects on protein stability and function.
• Genome-scale prediction: Can be applied to scan proteomes, exemplified by identifying 2,573 transcripts with predicted C-mannosylation sites in a human genome scan.

Methodology:

Compiled a dataset of 69 experimentally verified C-mannosylation sites, analyzed sequence context, examined Protein Data Bank structures for solvent exposure, and trained a neural network using a 21-residue input window with explicit encoding of WXXW motif presence/absence, optimizing model parameters to achieve high predictive accuracy.