A-Prot

A-Prot predicts protein 3D structures by leveraging MSA Transformer-derived features to infer residue-residue distances, dihedral angles, and long-range contacts for structural modeling.

Key Features:

• MSA Transformer integration: Uses MSA Transformer to extract evolutionary information from multiple sequence alignments (MSAs).
• Feature extraction: Extracts an MSA feature tensor and row attention maps for downstream prediction of residue-residue distances and dihedral angles.
• Long-range contact prediction: Predicts long-range contacts between residues to inform tertiary structure and spatial arrangement.
• Computational efficiency: Achieves high accuracy with relatively lower computational demands compared to more resource-intensive models.
• CASP14 validation: Demonstrated performance in CASP14 on free-modeling and hard template-based targets, outperforming most top server groups.

Scientific Applications:

• Protein function prediction: Infers 3D structures to support interpretation of protein functions, interactions, and mechanisms.
• Drug discovery and design: Provides precise structural models useful for identifying drug targets and designing protein-interacting molecules.
• Evolutionary studies: Leverages evolutionary signals in MSAs to study protein family trajectories and structural adaptations.

Methodology:

Extracts an MSA feature tensor and row attention maps from MSAs and uses them to predict residue-residue distances and dihedral angles that are assembled into 3D protein models.