AlphaFold 2

AlphaFold 2 predicts three-dimensional (3D) protein structures from amino acid sequences using deep learning to address the protein folding problem.

Key Features:

• Neural network-based approach: Utilizes advanced neural networks that integrate physical and biological knowledge to model protein structures.
• Integration of multi-sequence alignments: Incorporates multi-sequence alignments and evolutionary information into its deep learning framework to inform residue-residue relationships.
• Atomic-level accuracy: Produces predictions at atomic-level accuracy, including for proteins lacking homologous experimental structures.
• Validation and performance: Demonstrated high performance and was validated during CASP14 against experimental methods.

Scientific Applications:

• Structural bioinformatics: Provides accurate models for proteins without experimentally determined structures to support large-scale structural analyses.
• Mechanistic insights into protein function: Enables interpretation of molecular mechanisms relevant to drug discovery and enzyme engineering by supplying 3D structural models.
• Bridging experimental gaps: Alleviates structural-coverage bottlenecks by supplying rapid computational predictions that complement lengthy experimental structure determination.

Methodology:

Employs advanced neural networks that integrate multi-sequence alignments, evolutionary information, and physical and biological knowledge to predict atomic-resolution protein structures.