PCI-SS

PCI-SS predicts protein secondary structure by classifying residues as alpha-helices, beta-strands, or non-regular structures using Parallel Cascade Identification and PSI-BLAST-derived evolutionary profiles to inform nonlinear dynamic models.

Key Features:

• Three-state classification: Distinguishes alpha-helices, beta-strands, and non-regular structures at the residue level.
• Input profiles: Uses divergent evolutionary profiles produced by PSI-BLAST as primary input features.
• Nonlinear system modeling: Employs Parallel Cascade Identification (PCI) to construct black-box dynamic nonlinear systems aimed at simulating aspects of protein folding.
• Problem decomposition: Splits the three-state prediction into three binary sub-problems, each addressed by two layers of PCI classifiers.
• Parameter optimization: Applies genetic algorithms (GAs) to optimize PCI model parameters.
• Dataset construction: Uses datasets curated to eliminate homology between training and testing data.
• Benchmarking: Evaluated against nine methods on 125 protein chains dissimilar to training data.
• Ensemble integration: Combines with PSIPRED to improve prediction accuracy and reduce detrimental errors by 25%, improving BAD scores while maintaining Q3 error rates.
• File formats: Represents input protein sequences in XML and encodes predictions in machine-readable formats.

Scientific Applications:

• Residue-level secondary structure prediction: Predicts local secondary structure states (alpha-helix, beta-strand, non-regular) for protein sequences.
• Ensemble/consensus prediction: Integrates with PSIPRED to produce improved consensus predictions and reduce detrimental errors.
• Benchmarking on non-homologous sets: Assesses predictive performance on datasets constructed to be dissimilar from training data.
• Nonlinear folding-modeling studies: Uses black-box dynamic nonlinear systems to simulate aspects of protein folding dynamics.

Methodology:

PCI-SS uses PSI-BLAST-derived divergent evolutionary profiles as input to Parallel Cascade Identification (PCI) black-box dynamic nonlinear system models; the three-state problem is decomposed into three binary sub-problems solved by two-layer PCI classifiers whose parameters are optimized by genetic algorithms, and datasets are curated to remove homology between training and testing.