Near-native protein loop sampling using nonparametric density estimation accommodating sparcity.

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Protein Modeling

Background:

• Protein loop regions present significant challenges in template-based modeling (TBM) due to their high variability and limited homologous template data.
• Existing methods struggle to accurately model loop structures, impacting the overall accuracy of protein structure prediction.

Purpose of the Study:

• To develop a novel, knowledge-based method for enhanced protein loop sampling.
• To improve the accuracy and efficiency of modeling variable loop regions in protein structures.
• To leverage homologous torsion angle information for estimating accurate dihedral angle distributions.

Main Methods:

• A knowledge-based loop sampling method utilizing homologous torsion angle information.
• Estimation of continuous joint backbone dihedral angle probability density using Dirichlet process mixture of hidden Markov models (DPM-HMM).
• Rapid model generation from sampled distributions, followed by enrichment using an end-to-end distance filter.

Main Results:

• The DPM-HMM method achieved high accuracy, with candidates as low as 0.45 Å RMSD.
• It performs comparably or superiorly to template-based methods for canonical loops (e.g., immunoglobulin CDRs).
• The method effectively models loops in cases with poor or limited template availability, producing structures within 3.66 Å RMSD for loops up to 17 residues.

Conclusions:

• The DPM-HMM method offers a significant advantage in protein loop modeling by consistently sampling near-native structures.
• This automated approach successfully models canonical and non-canonical loops without specific template bias or manual intervention.
• The method demonstrates general applicability and effectiveness in realistic scenarios like the CASP9 experiment.