MUFOLD

MUFOLD clusters de novo protein tertiary structure models to select near‑native representatives using novel distance measures Dscore1 and Dscore2.

Key Features:

• Innovative distance measures: MUFOLD-CL computes Dscore1 and Dscore2 from protein distance matrices to assess model similarity in place of pairwise RMSD and TM-score calculations.
• High correlation with established metrics: Dscore1 shows high correlation with RMSD and Dscore2 shows high correlation with TM-score, preserving accuracy in identifying near‑native models.
• Linear time complexity: The Dscore1-based clustering algorithm scales linearly with the number of models, avoiding the quadratic time of pairwise comparisons.
• Enhanced cluster representative selection: MUFOLD-CL uses Dscore2 to select cluster representatives, improving representative quality with minimal extra computation.
• Fast data visualization: Dscore distribution–based visualization summarizes datasets of approximately 500,000 models in seconds to minutes.

Scientific Applications:

• De novo model selection: Selecting near‑native structures from large populations generated by de novo protein structure prediction.
• Structural biology analysis: Streamlining analysis of protein structure ensembles to study protein structure and dynamics.
• Downstream research workflows: Providing representative models to support drug discovery, enzyme design, and investigations of disease mechanisms.

Methodology:

Compute Dscore1 and Dscore2 from protein distance matrices, replace traditional pairwise RMSD/TM-score calculations with matrix-based comparisons, apply Dscore1-based linear-time clustering, use Dscore2 to select cluster representatives, and generate Dscore distribution visualizations for large datasets.