A-Prot: Protein structure modeling using MSA transformer

In this study, we propose a new protein 3D structure modeling method, A-Prot, using MSA Transformer, one of the state-of-the-art protein lan-guage models. For a given MSA, an MSA feature tensor and row attention maps are extracted and converted into 2D residue-residue distance and dihedral angle predictions. We demonstrated that A-Prot predicts long-range contacts better than the existing methods. Additionally, we modeled the 3D structures of the free modeling and hard template-based modeling targets of CASP14. The assessment shows that the A-Prot models are more accurate than most top server groups of CASP14. These results imply that A-Prot captures evolutionary and structural infor-mation of proteins accurately with relatively low computational cost. Thus, A-Prot can provide a clue for the development of other protein prop-erty prediction methods.

Epistasis at the SARS-CoV-2 RBD Interface and the Propitiously Boring Implications for Vaccine Escape

At the time of this writing, August 2021, potential emergence of vaccine escape variants of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is a grave global concern. The interface between the receptor-binding domain (RBD) of SARS-CoV-2 spike (S) protein and the host receptor (ACE2) overlap with the binding site of principal neutralizing antibodies (NAb), limiting the repertoire of viable mutations. Nonetheless, variants with multiple mutations in the RBD have rose to dominance. Non-additive, epistatic relationships among RBD mutations are apparent, and assessing the impact of such epistasis on the mutational landscape is crucial. Epistasis can substantially increase the risk of vaccine escape and cannot be completely characterized through the study of the wild type (WT) alone. We employed protein structure modeling using Rosetta to compare the effects of all single mutants at the RBD-NAb and RBD-ACE2 interfaces for the WT, Gamma (417T, 484K, 501Y), and Delta variants (452R, 478K). Overall, epistasis at the RBD surface appears to be limited and the effects of most multiple mutations are additive. Epistasis at the Delta variant interface weakly stabilizes NAb interaction relative to ACE2, whereas in the Gamma variant, epistasis more substantially destabilizes NAb interaction. These results suggest that the repertoire of potential escape mutations for the Delta variant is not substantially different from that of the WT, whereas Gamma poses a moderately greater risk for enhanced vaccine escape. Thus, the modest ensemble of mutations relative to the WT shown to reduce vaccine efficacy might constitute the majority of all possible escape mutations.SignificancePotential emergence of vaccine escape variants of SARS-CoV-2 is arguably the most pressing problem during the COVID-19 pandemic as vaccines are distributed worldwide. We employed a computational approach to assess the risk of antibody escape resulting from mutations in the receptor-binding domain of the spike protein of the wild type SARS-CoV-2 virus as well as the Gamma and Delta variants. The results indicate that emergence of escape mutants is somewhat less likely for the Delta variant than for the wild type and moderately more likely for the Gamma variant. We conclude that the small set of escape-enhancing mutations already identified for the wild type is likely to include the majority of all possible mutations with this effect, a welcome finding.

Diversity analysis of genes encoding Mfa1 fimbrial components in Porphyromonas gingivalis strains

Porphyromonas gingivalis, a gram-negative anaerobic bacterium, is associated with the development of periodontal disease. The genetic diversity in virulence factors, such as adhesive fimbriae, among its strains affects the bacterial pathogenicity. P. gingivalis generally expresses two distinct types of fimbriae, FimA and Mfa1. Although the genetic diversity of fimA, encoding the major FimA fimbrilin protein, has been characterized, the genes encoding the Mfa1 fimbrial components, including the Mfa1 to Mfa5 proteins, have not been fully studied. We, therefore, analyzed their genotypes in 12 uncharacterized and 62 known strains of P. gingivalis (74 strains in total). The mfa1 genotype was primarily classified into two genotypes, 53 and 70. Additionally, we found that genotype 70 could be further divided into two subtypes (70A and 70B). The diversity of mfa2 to mfa4 was consistent with the mfa1 genotype, although no subtype in genotype 70 was observed. Protein structure modeling showed high homology between the genotypes in Mfa1 to Mfa4. The mfa5 gene was classified into five genotypes (A to E) independent of other genotypes. Moreover, genotype A was further divided into two subtypes (A1 and A2). Surprisingly, some strains had two mfa5 genes, and the 2nd mfa5 exclusively occurred in genotype E. The Mfa5 protein in all genotypes showed a homologous C-terminal half, including the conserved C-terminal domain recognized by the type IX secretion system. Furthermore, the von Willebrand factor domain at the N-terminal was detected only in genotypes A to C. The mfa1 genotypes partially correlated with the ragA and ragB genotypes (located immediately downstream of the mfa gene cluster) but not with the fimA genotypes.

DeepEMhancer: a deep learning solution for cryo-EM volume post-processing

Cryo-EM maps are valuable sources of information for protein structure modeling. However, due to the loss of contrast at high frequencies, they generally need to be post-processed to improve their interpretability. Most popular approaches, based on global B-factor correction, suffer from limitations. For instance, they ignore the heterogeneity in the map local quality that reconstructions tend to exhibit. Aiming to overcome these problems, we present DeepEMhancer, a deep learning approach designed to perform automatic post-processing of cryo-EM maps. Trained on a dataset of pairs of experimental maps and maps sharpened using their respective atomic models, DeepEMhancer has learned how to post-process experimental maps performing masking-like and sharpening-like operations in a single step. DeepEMhancer was evaluated on a testing set of 20 different experimental maps, showing its ability to reduce noise levels and obtain more detailed versions of the experimental maps. Additionally, we illustrated the benefits of DeepEMhancer on the structure of the SARS-CoV-2 RNA polymerase.

CirPred, the first structure modeling and linker design system for circularly permuted proteins

Background This work aims to help develop new protein engineering techniques based on a structural rearrangement phenomenon called circular permutation (CP), equivalent to connecting the native termini of a protein followed by creating new termini at another site. Although CP has been applied in many fields, its implementation is still costly because of inevitable trials and errors. Results Here we present CirPred, a structure modeling and termini linker design method for circularly permuted proteins. Compared with state-of-the-art protein structure modeling methods, CirPred is the only one fully capable of both circularly-permuted modeling and traditional co-linear modeling. CirPred performs well when the permutant shares low sequence identity with the native protein and even when the permutant adopts a different conformation from the native protein because of three-dimensional (3D) domain swapping. Linker redesign experiments demonstrated that the linker design algorithm of CirPred achieved subangstrom accuracy. Conclusions The CirPred system is capable of (1) predicting the structure of circular permutants, (2) designing termini linkers, (3) performing traditional co-linear protein structure modeling, and (4) identifying the CP-induced occurrence of 3D domain swapping. This method is supposed helpful for broadening the application of CP, and its web server is available at http://10.life.nctu.edu.tw/CirPred/ and http://lo.life.nctu.edu.tw/CirPred/.

Localnet: A Simple Recurrent Neural Network Model for Protein Secondary Structure Prediction Using Local Amino Acid Sequences Only

BackgroundProtein secondary structure prediction (PSSP) is important for protein structure modeling and design. Over the past a few years, deep learning models have shown promising results for PSSP. However, the current good performers for PSSP often require evolutionary information such as multiple sequence alignments and even real protein structures (templates), entire protein sequences, and amino acid property profiles. ResultsIn this study, we used a fixed-size window of adjacent residues and only amino acid sequences, without any evolutionary information, as inputs, and developed a very simple, yet accurate RNN model: LocalNet. The accuracy for three states of secondary structures is as high as 85.15%, indicating that the local amino acid sequence itself contains enough information for PSSP, a well-known classical view. By comparing to other predictors, we also achieve an state-of-art accuracy on dataset of CASP11, CASP12 and CASP13.ConclusionThe well-trained models are expected to have good applications in protein structure modeling and protein design. This model can be downloaded from https://github.com/lake-chao/protein-secondary-structure-prediction.