Large-scale structure prediction by improved contact predictions and model quality assessment

Area of Science:

• Structural Biology
• Computational Biology
• Bioinformatics

Background:

• Accurate protein contact predictions are crucial for determining protein structures.
• Recent advancements enable accurate contact map predictions for smaller protein families.
• The utility of these predictions for modeling smaller families remains largely unexplored.

Purpose of the Study:

• To develop and evaluate a pipeline for predicting protein structures using contact predictions.
• To assess the impact of model quality estimation on reliable structure identification.
• To apply the pipeline to predict structures for a large set of Pfam families with unknown structures.

Main Methods:

• The PconsFold2 pipeline integrates contact predictions (PconsC3), protein folding (CONFOLD), and model quality estimation.
• Statistical methods and machine learning were employed for contact prediction and quality assessment.
• The pipeline was applied to 6379 Pfam families lacking known structures.

Main Results:

• PconsFold2 significantly enhances the identification of reliable protein models through quality estimation.
• The pipeline successfully predicted structures for up to 558 Pfam families with 90% specificity.
• This includes 415 novel protein structures not previously reported.

Conclusions:

• PconsFold2 is an effective tool for predicting protein structures from contact predictions, particularly for smaller families.
• The pipeline expands the repertoire of known protein structures, aiding further biological research.
• The availability of datasets and models facilitates reproducibility and future studies.