MetaPSICOV

MetaPSICOV predicts residue-residue contacts from multiple sequence alignments to improve the accuracy of de novo 3D protein structure modeling by integrating covariation signals and sequence-derived features.

Key Features:

• Integration of Multiple Approaches: MetaPSICOV combines three distinct covariation inference methods applied to multiple sequence alignments to capture complementary evolutionary signals.
• Incorporation of Sequence-Derived Features: It incorporates extensive sequence-derived features and metrics assessing both local and global qualities of the input multiple sequence alignment.
• Two-Stage Prediction Framework: It employs a two-stage predictive framework where initial contact predictions are generated in stage one and refined by filtering in stage two.
• Hydrogen Bond Network Prediction (MetaPSICOV-HB): MetaPSICOV-HB predicts long-range hydrogen bond networks with donor and acceptor assignments and attains a precision of 0.69 for the top-L/10 predicted hydrogen bonds.
• Benchmark Performance: On the original PSICOV benchmark set of 150 protein families it achieves a mean precision of 0.54 for top-L predicted long-range contacts, approximately 60% improvement over PSICOV and around 40% better than CCMpred.

Scientific Applications:

• 3D Protein Modeling: Improved contact predictions aid de novo protein structure prediction and, when used with FRAGFOLD, increase median TM-scores of models by 0.05 compared to PSICOV.

Methodology:

MetaPSICOV integrates three covariation inference methods on multiple sequence alignments, combines sequence-derived features and MSA-quality metrics, applies a two-stage prediction and filtering process, and predicts long-range hydrogen-bond networks with donor and acceptor assignment.