Fast and accurate Ab Initio Protein structure prediction using deep learning potentials

Despite the immense progress recently witnessed in protein structure prediction, the modeling accuracy for proteins that lack sequence and/or structure homologs remains to be improved. We developed an open-source program, DeepFold, which integrates spatial restraints predicted by multi-task deep residual neural-networks along with a knowledgebased energy function to guide its gradient-descent folding simulations. The results on large-scale benchmark tests showed that DeepFold creates full-length models with accuracy significantly beyond classical folding approaches and other leading deep learning methods. Of particular interest is the modeling performance on the most difficult targets with very few homologous sequences, where DeepFold achieved an average TM-score that was 40.3% higher than trRosetta and 44.9% higher than DMPfold. Furthermore, the folding simulations for DeepFold were 262 times faster than traditional fragment assembly simulations. These results demonstrate the power of accurately predicted deep learning potentials to improve both the accuracy and speed of ab initio protein structure prediction. Author summary Template-free protein structure prediction remains an important unsolved problem. We proposed a new pipeline to construct full-length protein structures by coupling multiplelevel deep learning potentials with fast gradient-based folding simulations. The large-scale benchmark tests demonstrated significant advantages in both accuracy and speed over other fragment-assembly and deep learning-based approaches. The results revealed that the key factor for the success of the deep learning approach is its ability to provide an abundant set of accurate spatial restraints (similar to 93*L where L is the protein length), which help smooth the energy landscape and make gradient-based simulation searching a feasible optimization tool. Nevertheless, extensive folding simulations are still needed for the cases where only sparse restraints are available as provided by threading alignments and low-resolution structural biology experiments.