Accurate prediction of protein structures and interactions using a three-track neural network
Deep learning supports protein folding
In 1972, Anfinsen won a Nobel Prize for demonstrating a link between a protein’s amino acid sequence and its three-dimensional structure. Since 1994, scientists have participated in the biannual Structure Prediction Critical Appraisal Protein Folding Challenge (CASP). Deep learning methods took center stage at CASP14, with DeepMind’s Alphafold2 achieving remarkable precision. Baek et al. explored network architectures based on the DeepMind framework. They used a three-track network to simultaneously process sequence, distance, and coordination information and obtained accuracies close to that of DeepMind. The method, RoseTTA fold, can solve complex X-ray crystallography and cryo-electron microscopy problems and generate precise models of protein-protein complexes.
Science, abj8754, this issue p. 871
DeepMind presented particularly accurate predictions at the recent 14th Critical Assessment of Structure Prediction (CASP14) conference. We have explored network architectures that incorporate related ideas and achieved the best performance with a three-track network in which information at the one-dimensional (1D) sequence level, 2D distance map level, and coordinate level 3D are successively transformed and integrated. The three-track array produces structure predictions with accuracies approaching those of DeepMind in CASP14, enables complex X-ray crystallography and cryo-electron microscopy structure modeling problems to be quickly solved, and provides information on functions. proteins of currently unknown structure. The network also enables the rapid generation of precise protein-protein complex models from sequence information alone, bypassing traditional approaches that require modeling of individual subunits followed by docking. We make the method available to the scientific community to accelerate biological research.