pep2d

pep2d predicts peptide secondary structure to assign residue-level helix, beta-sheet and coil states for peptides, supporting peptide structural analysis and design.

Key Features:

• Dataset Utilization: Models were trained, tested, and evaluated on a dataset of 3,107 unique peptides.
• Secondary Structure Classes: Predicts helix, beta-sheet and coil states for peptides.
• Machine Learning (binary profiles): Initial models used binary profiles and achieved an overall Q3 accuracy of 79.5%.
• Evolutionary Information (PSSM): Incorporation of Position-Specific Scoring Matrix (PSSM) profiles increased overall accuracy to 83.5%.
• Benchmarking against PSIPRED: Compared to PSIPRED (overall maximum accuracy 76.9%), PEP2D performed better on small peptides (<10 residues) where PSIPRED reached Q3 of 71.4%.
• Per-class performance: PEP2D reported Q3 accuracies of 74% for beta-sheets and 87% for coils, compared with PSIPRED's 54.4% for beta-sheets and 77.9% for coils.
• Segment Overlap (SOV): Evaluated by SOV, PEP2D achieved 76.7 versus PSIPRED's 69.3, indicating improved segment-level agreement.

Scientific Applications:

• Therapeutic peptide design: Provides secondary-structure information to support drug design and development of therapeutic peptides.
• Peptide structure–function analysis: Aids in elucidating peptide functional mechanisms by improving prediction of secondary-structure elements.

Methodology:

Models were developed using machine-learning classifiers trained on binary profiles of 3,107 peptides, enhanced by PSSM-derived evolutionary profiles; performance was quantified by Q3 accuracy and segment overlap (SOV) and benchmarked against PSIPRED.