Structpred is a collection of webservers for protein structure prediction.
We will provide several tools for users to make prediction at various levels.
GANProDist is a lightweight GAN-based protein inter-residue distance prediction web server. Unlike mainstream protein distance predictors that utilize neural networks for multi-classification tasks to predict discrete probability distribution of different distance bins, GANProDist is capable to capture the delicate geometric relationship between residue pairs and thus could predict the continuous, real-valued distance rapidly and satisfactorily.
AmoebaContact is a web server that predicts the residue contact maps for a target protein at multiple contact cutoffs. Unlike mainstream contact predictors that utilize human-designed neural networks, AmoebaContact adopts a set of network architectures that are found as optimal for contact prediction through automatic searching and predicts the residue contacts at a series of cutoffs. We also provide a contact-assisted folder GDFold for rapid protein structure prediction. Combination of AmoebaContact and GDFold allows quick reconstruction of the protein structure, with comparable model quality to the state-of-the-art protein structure prediction methods.
DeepFragLib is a web server to build the fragment library for a target protein sequence, which could be used as input for subsequent fragment-assembly-based protein structure prediction. DeepFragLib improves the identification of good template fragments with low sequence homology, by engaging multiple deep learning techniques. Specifically, the algorithm combines a number of classification models using BiLSTM neural networks and one regression model using ultra-deep ResNeXt with the cyclically dilated convolution. On the CASP11-CASP13 test sets, DeepFragLib significantly outperforms the other state-of-the-art algorithms in respect to the precision of identified fragments. Moreover, the fragment library constructed thereby could effectively improve the quality of sampled structure decoys when integrated with ROSETTA.
RDb2C2 is a web server to predict the residue pairing between interacting β strands of a protein given the amino acid sequence. Knowledge of the β-β pairing information could facilitate the structure modeling of mainly β proteins. RDb2C2 adopts the ridge detection method to infer the characteristics of interacting β strands and then utilizes the residual neural network (ResNet) to improve the prediction of β-β residue pairing. This algorithm could be ranked as one of the best methods for the β-β contact prediction, based on our benchmark test on BetaSheet916 and BetaSheet1452 sets.
DeepConPred2 is a web server that predicts the residue contact map of a target protein from the amino acid sequence. It has 3 modules for different steps of contact prediction and works well on proteins even with limited numbers of homologous sequences. DeepConPred2 could be ranked as one of the state-of-the-art methods in contact prediction according to our tests on different testing sets including the CASP11 and CASP12 sets.