SPICKER

SPICKER identifies near-native protein structures from large ensembles of decoy models generated during computational simulations.

Key Features:

• Clustering Strategy: Organizes protein structure decoys into clusters based on structural similarity and prioritizes densely populated clusters as likely near-native folds.
• Application Scope: Validated on 1,489 benchmark proteins representative of the Protein Data Bank (PDB) at ≤35% sequence identity, with up to 280,000 decoy structures per target.
• Performance Metrics: Identifies folds that rank within the top 1.4% of all decoys by RMSD, and for 78% of proteins the RMSD difference between identified models and the best individual decoy is <1 Å.
• Convergence and Improvement: Effectively handles targets with converged conformational distributions and shows improvement over previous clustering algorithms in identifying native-like structures.

Scientific Applications:

• Protein structure prediction: Selects near-native models from large decoy ensembles to support prediction workflows.
• Protein folding studies: Analyzes conformational landscapes derived from simulation decoys to study folding behavior.
• Large-scale decoy analysis: Applies to high-throughput simulation outputs, accommodating up to hundreds of thousands of decoy structures per target.
• Benchmarking and method evaluation: Provides RMSD-based performance metrics for benchmarking structure prediction and clustering methods.

Methodology:

Clusters decoy models by structural similarity and ranks clusters by population density to identify near-native conformations, implemented within the TASSER framework.