MOTIVATION: Predicting protein structures with high accuracy is a critical challenge for the broad community of life sciences and industry. Despite progress made by deep neural networks like AlphaFold2, there is a need for further improvements in the quality of detailed structures, such as side-chains, along with protein backbone structures. RESULTS: Building upon the successes of AlphaFold2, the modifications we made include changing the losses of side-chain torsion angles and frame aligned point error, adding loss functions for side chain confidence and secondary structure prediction, and replacing template feature generation with a new alignment method based on conditional random fields. We also performed re-optimization by conformational space annealing using a molecular mechanics energy function which integrates the potential energies obtained from distogram and side-chain prediction. In the CASP15 blind test for single protein and domain modeling (109 domains), DeepFold ranked fourth among 132 groups with improvements in the details of the structure in terms of backbone, side-chain, and Molprobity. In terms of protein backbone accuracy, DeepFold achieved a median GDT-TS score of 88.64 compared with 85.88 of AlphaFold2. For TBM-easy/hard targets, DeepFold ranked at the top based on Z-scores for GDT-TS. This shows its practical value to the structural biology community, which demands highly accurate structures. In addition, a thorough analysis of 55 domains from 39 targets with publicly available structures indicates that DeepFold shows superior side-chain accuracy and Molprobity scores among the top-performing groups. AVAILABILITY AND IMPLEMENTATION: DeepFold tools are open-source software available at https://github.com/newtonjoo/deepfold.

Protein folding rate is one of the important properties of a protein. Protein folding rate prediction is useful for understanding protein folding process and guiding protein design. In this study, we developed a support vector machine (SVM) based method to predict protein folding kinetic types (two-state or non-two-state) and the real-value folding rate using the features calculated from the three-dimensional structure such as contact order, various properties from the non-local contact clusters, secondary structural information and sequence length. We systematically studied the contributions of individual features to folding rate prediction. Based on the highest contributions of individual features, we trained our machine using leave one out cross-validation and tested on a testing dataset. The Pearson correlation coefficient, mean absolute difference and root mean square error between the predicted and experimental folding rates (base-10 logarithmic scale) are 0.814, 0.752 and 0.910 for two-state proteins, and 0.860, 0.687 and 0.876 for non-two-state proteins. Moreover, our method predicts whether a protein of known atomic structure folds according to two-state or non-two-state kinetics and correctly classifies 80% of the folding mechanism on a testing dataset. Finally, we evaluated the performance of our method along with the other eight existing protein folding rate prediction tools on non-overlapping benchmarking dataset. The prediction performance will also be reported and discussed.

BACKGROUND: In template-based modeling when using a single template, inter-atomic distances of an unknown protein structure are assumed to be distributed by Gaussian probability density functions, whose center peaks are located at the distances between corresponding atoms in the template structure. The width of the Gaussian distribution, the variability of a spatial restraint, is closely related to the reliability of the restraint information extracted from a template, and it should be accurately estimated for successful template-based protein structure modeling. RESULTS: To predict the variability of the spatial restraints in template-based modeling, we have devised a prediction model, Sigma-RF, by using the random forest (RF) algorithm. The benchmark results on 22 CASP9 targets show that the variability values from Sigma-RF are of higher correlations with the true distance deviation than those from Modeller. We assessed the effect of new sigma values by performing the single-domain homology modeling of 22 CASP9 targets and 24 CASP10 targets. For most of the targets tested, we could obtain more accurate 3D models from the identical alignments by using the Sigma-RF results than by using Modeller ones. CONCLUSIONS: We find that the average alignment quality of residues located between and at two aligned residues, quasi-local information, is the most contributing factor, by investigating the importance of input features used in the RF machine learning. This average alignment quality is shown to be more important than the previously identified quantity of a local information: the product of alignment qualities at two aligned residues.