MUFOLD_CL

MUFOLD_CL clusters protein structure models using two distance-matrix-derived metrics, Dscore1 and Dscore2, to enable efficient clustering and selection of near-native structures.

Key Features:

• Innovative Distance Measures: Dscore1 and Dscore2 are derived from protein distance matrices, with Dscore1 evaluating model differences and Dscore2 evaluating model similarities.
• High Correlation with Established Metrics: Dscore1 shows strong correlation with RMSD and Dscore2 shows strong correlation with TM-score, maintaining consistency with established metrics.
• Linear Time Complexity: Utilizing Dscore1 for clustering reduces algorithmic complexity to linear time with respect to the number of models.
• Enhanced Cluster Representation Quality: Dscore2 is used to select representative models from clusters, improving representative quality with minimal additional computation.
• Fast Data Visualization: Dscore distributions enable rapid visualization of very large model sets (e.g., ~500,000 models) to assess clustering outcomes.

Scientific Applications:

• Structural model management and quality assessment: Supports management and interpretation of large collections of protein structure models generated by de novo prediction methods and aids identification of near-native structures for studies of protein function and therapeutic design.

Methodology:

Computes Dscore1 and Dscore2 from protein distance matrices and compares models via these measures instead of traditional pairwise RMSD/TM-score calculations to reduce computational demands.