DMPfold

DMPfold predicts de novo protein structures by using deep learning to infer inter-atomic distance bounds, main-chain hydrogen bond networks, and torsion angles for iterative structural modelling and genome-scale structural annotation.

Key Features:

• Deep learning predictions: Employs deep learning algorithms to predict inter-atomic distance bounds, main-chain hydrogen bond networks, and torsion angles.
• Iterative modelling: Builds protein models iteratively, refining structural constraints with each iteration.
• Addresses covariation limitations: Overcomes limitations of amino acid covariation methods for small protein families.
• Comparative performance: Demonstrated superior performance on CASP12 domains compared with two established methods and shows effectiveness for transmembrane proteins.
• Pfam dark families: Generated confident models for 25% of Pfam domains lacking known structures within a week on a 200-core computing cluster.
• Human proteome coverage: Produced accurate models for 16% of human proteome UniProt entries lacking structural data, including cases with fewer than 100 sequences.

Scientific Applications:

• Structural biology: Enables de novo structural modelling to support interpretation of protein function and interactions.
• Genome annotation: Provides structural models for Pfam domains and other protein families to assist genome-scale structural annotation.
• Dark family characterization: Generates models for "dark families" lacking experimental structures to facilitate hypothesis generation about function.
• Transmembrane protein modelling: Applicable to modelling transmembrane proteins where accurate inter-atomic constraints improve model quality.
• Human proteome structural coverage: Expands structural information for UniProt entries lacking experimental structures, including proteins with limited sequence homologs.

Methodology:

Uses deep learning to predict inter-atomic distance bounds, main-chain hydrogen bond networks, and torsion angles, and constructs models iteratively while refining structural constraints.