Evaluation of Deep Neural Network ProSPr for Accurate Protein Distance Predictions on CASP14 Targets

The field of protein structure prediction has recently been revolutionized through the introduction of deep learning. The current state-of-the-art tool AlphaFold2 can predict highly accurate structures; however, it has a prohibitively long inference time for applications that require the folding of hundreds of sequences. The prediction of protein structure annotations, such as amino acid distances, can be achieved at a higher speed with existing tools, such as the ProSPr network. Here, we report on important updates to the ProSPr network, its performance in the recent Critical Assessment of Techniques for Protein Structure Prediction (CASP14) competition, and an evaluation of its accuracy dependency on sequence length and multiple sequence alignment depth. We also provide a detailed description of the architecture and the training process, accompanied by reusable code. This work is anticipated to provide a solid foundation for the further development of protein distance prediction tools.

Evaluation of Deep Neural Network ProSPr for Accurate Protein Distance Predictions on CASP14 Targets

The field of protein structure prediction has recently been revolutionized through the introduction of deep learning. The current state-of-the-art tool AlphaFold2 can predict highly accurate structures, however, it has a prohibitively long inference time for applications that require the folding of hundreds of sequences. The prediction of protein structure annotations, such as amino acid distances, can be achieved at a higher speed with existing tools, such as the ProSPr network. Here, we report on important updates to the ProSPr network, its performance on the recent Critical Assessment of Structure Prediction (CASP14) competition, and an evaluation of its accuracy dependency on multiple sequence alignment depth. We also provide a detailed description of the architecture and the training process, accompanied by reusable code. This work is anticipated to provide a solid foundation for the further development of protein distance prediction tools.

Improving deep learning-based protein distance prediction in CASP14

Accurate prediction of residue-residue distances is important for protein structure prediction. We developed several protein distance predictors based on a deep learning distance prediction method and blindly tested them in the 14th Critical Assessment of Protein Structure Prediction (CASP14). The prediction method uses deep residual neural networks with the channel-wise attention mechanism to classify the distance between every two residues into multiple distance intervals. The input features for the deep learning method include co-evolutionary features as well as other sequence-based features derived from multiple sequence alignments (MSAs). Three alignment methods are used with multiple protein sequence/profile databases to generate MSAs for input feature generation. Based on different configurations and training strategies of the deep learning method, five MULTICOM distance predictors were created to participate in the CASP14 experiment. Benchmarked on 37 hard CASP14 domains, the best performing MULTICOM predictor is ranked 5th out of 30 automated CASP14 distance prediction servers in terms of precision of top L/5 long-range contact predictions (i.e. classifying distances between two residues into two categories: in contact (< 8 Angstrom) and not in contact otherwise) and performs better than the best CASP13 distance prediction method. The best performing MULTICOM predictor is also ranked 6th among automated server predictors in classifying inter-residue distances into 10 distance intervals defined by CASP14 according to the F1 measure. The results show that the quality and depth of MSAs depend on alignment methods and sequence databases and have a significant impact on the accuracy of distance prediction. Using larger training datasets and multiple complementary features improves prediction accuracy. However, the number of effective sequences in MSAs is only a weak indicator of the quality of MSAs and the accuracy of predicted distance maps. In contrast, there is a strong correlation between the accuracy of contact/distance predictions and the average probability of the predicted contacts, which can therefore be more effectively used to estimate the confidence of distance predictions and select predicted distance maps.

REALDIST: Real-valued protein distance prediction

Protein structure prediction continues to stand as an unsolved problem in bioinformatics and biomedicine. Deep learning algorithms and the availability of metagenomic sequences have led to the development of new approaches to predict inter-residue distances—the key intermediate step. Different from the recently successful methods which frame the problem as a multi-class classification problem, this article introduces a real-valued distance prediction method REALDIST. Using a representative set of 43 thousand protein chains, a variant of deep ResNet is trained to predict real-valued distance maps. The contacts derived from the real-valued distance maps predicted by this method, on the most difficult CASP13 free-modeling protein datasets, demonstrate a long-range top-L precision of 52%, which is 17% higher than the top CASP13 predictor Raptor-X and slightly higher than the more recent trRosetta method. Similar improvements are observed on the CAMEO ‘hard’ and ‘very hard’ datasets. Three-dimensional (3D) structure prediction guided by real-valued distances reveals that for short proteins the mean accuracy of the 3D models is slightly higher than the top human predictor AlphaFold and server predictor Quark in the CASP13 competition.

A fully open-source framework for deep learning protein real-valued distances

As deep learning algorithms drive the progress in protein structure prediction, a lot remains to be studied at this emerging crossway of deep learning and protein structure prediction. Recent findings show that inter-residue distance prediction, a more granular version of the well-known contact prediction problem, is a key to predict accurate models. We believe that deep learning methods that predict these distances are still at infancy. To advance these methods and develop other novel methods, we need a small and representative dataset packaged for fast development and testing. In this work, we introduce Protein Distance Net (PDNET), a dataset derived from the widely used DeepCov dataset and consists of 3456 representative protein chains for training and validation. It is packaged with all the scripts that were used to curate the dataset, generate the input features and distance maps, and scripts with deep learning models to train, validate and test. Deep learning models can also be trained and tested in a web browser using free platforms such as Google Colab. We discuss how this dataset can be used to predict contacts, distance intervals, and real-valued distances (in Å) by designing regression models. All scripts, training data, deep learning code for training, validation, and testing, and Python notebooks are available at https://github.com/ba-lab/pdnet/.

Comparative effects of oncogenic mutations G12C, G12V, G13D, and Q61H on local conformations and dynamics of K-Ras

ABSTRACTK-Ras is the most frequently mutated protein in human cancers. However, until very recently, its oncogenic mutants were viewed as undruggable. To develop inhibitors that directly target oncogenic K-Ras mutants, we need to understand both their mutant-specific and pan-mutant dynamics and conformations. Recently, we have investigated how the most frequently observed K-Ras mutation in cancer patients, G12D, changes its local dynamics and conformations1. Here, we extend our analysis to study and compare the local effects of other frequently observed oncogenic mutations, G12C, G12V, G13D and Q61H. For this purpose, we have performed Molecular Dynamics (MD) simulations of each mutant when active (GTP-bound) and inactive (GDP-bound), analyzed their trajectories, and compared how each mutant changes local residue conformations, inter-protein distance distributions, local flexibility and residue pair correlated motions. Our results reveal that in the four active oncogenic mutants we have studied, the α2 helix moves closer to the C-terminal of the α3 helix. However, P-loop mutations cause α3 helix to move away from Loop7, and only G12 mutations change the local conformational state populations of the protein. Furthermore, the motions of coupled residues are mutant-specific: G12 mutations lead to new negative correlations between residue motions, while Q61H destroys them. Overall, our findings on the local conformational states and protein dynamics of oncogenic K-Ras mutants can provide insights for both mutant-selective and pan-mutant targeted inhibition efforts.

ProSPr: Democratized Implementation of Alphafold Protein Distance Prediction Network

Deep neural networks have recently enabled spectacular progress in predicting protein structures, as demonstrated by DeepMin’s winning entry with Alphalfold at the latest Critical Assessment, of Structure Prediction competition (CASP13). The best protein prediction pipeline leverages intermolecular distance predictions to assemble a final protein model, but this distance prediction network has not been published. Here, we make a trained implementation of this network available to the broader scientific community. We also benchmark its predictive power in the related task of contact prediction against the CASP13 contact prediction winner TripletRes. Access to ProSPr will enable other labs to build on best in class protein distance predictions and to engineer superior protein reconstruction methods.

Hydrodynamic trapping measures the interaction between membrane-associated molecules

How membrane proteins distribute and behave on the surface of cells is determined by the molecules’ interaction potential. However, measuring this potential, and how it varies with protein-to-protein distance, has been challenging. We here present how a method we call hydrodynamic trapping can achieve this. Our method uses the focused liquid flow from a micropipette to locally accumulate molecules protruding from a lipid membrane. The interaction potential, as well as information about the dimensions of the studied molecule, are obtained by relating the degree of accumulation to the strength of the trap. We have used this to study four representative proteins, with different height-to-width ratios and protein properties; from the globular streptavidin, to the rod-like immune cell proteins CD2, CD4 and CD45. The obtained data illustrates how protein shape, glycosylation and flexibility influence the behaviour of membrane proteins as well as underline the general applicability of the method.