SELECTpro

SELECTpro selects the most native-like protein tertiary structure model from a set of predicted models using a structure-based energy function that integrates physical, statistical, and predicted-structural terms.

Key Features:

• Energy Function Composition: An energy function integrates physical and statistical terms with predicted-structural inputs, including predicted secondary structure, predicted solvent accessibility, predicted contact maps, beta-strand pairing, and side-chain hydrogen bonding.
• Independent Model Scoring: Models are scored independently rather than using consensus or redundancy-based ranking approaches.
• BLUNDER Metric: Implements the BLUNDER metric to mitigate effects of model redundancy and cases where poorer models receive better rankings than the most native-like model.
• CASP7 Performance: Participated in CASP7 new model quality assessment for 95 targets with an average GDT-TS difference of 5.07 between SELECTpro-ranked and most native-like models, with differences <1% for 18 targets and <10% for 66 targets.
• Ranking Accuracy: Ranked the single most native-like model first for 15 targets, within the top five for 39 targets, and within the top ten for 53 targets.
• Benchmark Performance: Outperformed I-TASSER on a benchmark of 16 small proteins with decoy sets of 12,500–20,000 models per protein.

Scientific Applications:

• Analogous fold recognition: Applied to analogous fold recognition by selecting native-like models among fold candidates.
• Refinement of sequence-structure alignments: Used to refine sequence-structure alignments through model ranking.
• De novo prediction: Applied to de novo structure prediction by ranking decoy models to identify native-like conformations.

Methodology:

Computes an energy score that integrates physical, statistical, and predicted-structural terms (predicted secondary structure, solvent accessibility, contact maps, beta-strand pairing, side-chain hydrogen bonding), scores models independently, and computes the BLUNDER metric.