AlphaFold is a deep learning model that achieves highly accurate prediction of protein structures. The model was validated in the 14th Critical Assessment of protein Structure Prediction (CASP14), where it outperformed other methods in accuracy. AlphaFold uses a novel neural network architecture that incorporates physical and biological knowledge about protein structures, leveraging multi-sequence alignments to design the deep learning algorithm. The model can predict protein structures with atomic accuracy even when no similar structure is known. AlphaFold's performance was tested on a large dataset of recently released PDB structures, demonstrating high accuracy and reliability. The model is scalable to very long proteins and can provide precise, per-residue estimates of its reliability. The AlphaFold network uses a combination of evolutionary, physical, and geometric constraints to improve the accuracy of structure prediction. The model was trained using both labeled and unlabeled data, with a self-distillation approach that enhances accuracy. AlphaFold has demonstrated its utility in the experimental community, both for molecular replacement and for interpreting cryogenic electron microscopy maps. The development of AlphaFold has the potential to revolutionize structural bioinformatics by enabling large-scale structural predictions. The model's success highlights the importance of integrating biological knowledge with deep learning techniques to improve the accuracy of protein structure prediction.AlphaFold is a deep learning model that achieves highly accurate prediction of protein structures. The model was validated in the 14th Critical Assessment of protein Structure Prediction (CASP14), where it outperformed other methods in accuracy. AlphaFold uses a novel neural network architecture that incorporates physical and biological knowledge about protein structures, leveraging multi-sequence alignments to design the deep learning algorithm. The model can predict protein structures with atomic accuracy even when no similar structure is known. AlphaFold's performance was tested on a large dataset of recently released PDB structures, demonstrating high accuracy and reliability. The model is scalable to very long proteins and can provide precise, per-residue estimates of its reliability. The AlphaFold network uses a combination of evolutionary, physical, and geometric constraints to improve the accuracy of structure prediction. The model was trained using both labeled and unlabeled data, with a self-distillation approach that enhances accuracy. AlphaFold has demonstrated its utility in the experimental community, both for molecular replacement and for interpreting cryogenic electron microscopy maps. The development of AlphaFold has the potential to revolutionize structural bioinformatics by enabling large-scale structural predictions. The model's success highlights the importance of integrating biological knowledge with deep learning techniques to improve the accuracy of protein structure prediction.