AlphaFold2

AlphaFold2 is a deep learning system developed by DeepMind that predicts the 3D structure of proteins from their amino acid sequence. It was first publicly unveiled at the CASP14 (Critical Assessment of protein Structure Prediction) competition in late 2020, where it achieved unprecedented accuracy, significantly outperforming all other methods. The system's success was attributed to its innovative architecture, which combines ideas from neural networks, attention mechanisms, and evolutionary information.

The core of AlphaFold2's approach involves an end-to-end neural network that takes an amino acid sequence as input and outputs a predicted 3D structure. It leverages multiple sequence alignments, which represent evolutionary relatedness between different proteins, to infer structural constraints. The system then uses a novel "invariant point attention" module to iteratively refine the protein structure, considering the relative positions and orientations of amino acid residues. This iterative process allows AlphaFold2 to build highly accurate models, often comparable to experimentally determined structures.

Following its success at CASP14, DeepMind released the AlphaFold2 code and a database of predicted protein structures, making this powerful tool accessible to the wider scientific community. This has led to a surge in protein structure prediction capabilities across various fields, including drug discovery, disease research, and fundamental biology. Scientists can now readily obtain reliable structural information for proteins that were previously difficult or impossible to study experimentally. The impact of AlphaFold2 has been transformative, accelerating research and opening up new avenues for scientific exploration.