In-silico prediction of protein biophysical traits is often hindered by the limited availability of experimental data and their heterogeneity. Training on limited data can lead to overfitting and poor generalizability to sequences distant from those in the training set. Additionally, inadequate use of scarce and disparate data can introduce biases during evaluation, leading to unreliable model performances being reported. Here, we present a comprehensive study exploring various approaches for protein fitness prediction from limited data, leveraging pre-trained embeddings, repeated stratified nested cross-validation, and ensemble learning to ensure an unbiased assessment of the performances. We applied our framework to introduce NanoMelt, a predictor of nanobody thermostability trained with a dataset of 640 measurements of apparent melting temperature, obtained by integrating data from the literature with 129 new measurements from this study. We find that an ensemble model stacking multiple regression using diverse sequence embeddings achieves state-of-the-art accuracy in predicting nanobody thermostability. We further demonstrate NanoMelt's potential to streamline nanobody development by guiding the selection of highly stable nanobodies. We make the curated dataset of nanobody thermostability freely available and NanoMelt accessible as a downloadable software and webserver.

The accumulation of polyethylene terephthalate (PET) waste has caused significant environmental pollution. Although biological depolymerization offers a promising solution, its efficiency remains constrained by the limited activity of PET-degrading enzymes. In this study, we designed a Structure-guided Loop-focused Iterative Mutagenesis (SLIM) strategy and rationally engineered the PET degradation enzyme ICCG for higher activity. The strategy was designed by demonstrating the critical role of the β8-α6 loop in type Ⅰ enzymes, which has currently not been reported. The best variant obtained, YITA (H183Y/L202I/I208T/T153A), exhibited 4.46-fold higher hydrolytic activity on amorphous PET at 72 °C compared to ICCG, outperforming other PET hydrolases, and exhibited superior degradation activity on real substrates such as cake containers and PET fibers. Conformational analysis revealed the key role of the remodeled β8-α6 loop in substrate binding and overall stability. Collectively, this study explores a promising approach to modifying PET hydrolase and lays a theoretical foundation for advancing bio-circular economy.

Tunnel engineering targets the access tunnels in enzymes, which is crucial for substrate binding and product release. Modifying the tunnels can lead to better biomass-degrading abilities of the lignocellulolytic enzymes. In this report, we have engineered the thermostable GH7 family endoglucanase from Aspergillus fumigatus (AfEgl7). The residues in the open tunnel having the highest bottleneck radius are mutated. Mutations are created (T229F, W356C) in the non-conserved region. The mutant W356C showed a 2-fold increase in product release rate (Vmax = 375.8 µM/min) and 2.5-fold higher catalytic activity (Kcat = 75.1 min-1) compared to wild-type (Vmax= 232 µM/min; Kcat = 30.9 min-1) using CM cellulose as substrate. The mutant T229F lost both catalytic activity and thermostability. Molecular dynamic simulations and docking studies of W356C revealed a change in structure near the product exit region, which may facilitate faster product release and account for the increased catalytic efficiency of the mutant. This study showed how redesigning the access pathways can be a promising strategy for protein engineering and de novo protein design by tailoring the open tunnel geometry to a ligand-specific manner.

The complex genotype-to-phenotype relationships in Mendelian diseases can be elucidated by mutation-induced disturbances to the networks of molecular interactions (interactomes) in human cells. Missense and nonsense mutations cause distinct perturbations within the human protein interactome, leading to functional and phenotypic effects with varying degrees of severity. Here, we structurally resolve the human protein interactome at atomic-level resolutions and perform structural and thermodynamic calculations to assess the biophysical implications of these mutations. We focus on a specific type of missense mutation, known as "quasi-null" mutations, which destabilize proteins and cause similar functional consequences (node removal) to nonsense mutations. We propose a "fold difference" quantification of deleteriousness, which measures the ratio between the fractions of node-removal mutations in datasets of Mendelian disease-causing and non-pathogenic mutations. We estimate the fold differences of node-removal mutations to range from 3 (for quasi-null mutations with folding ΔΔG ≥2 kcal/mol) to 20 (for nonsense mutations). We observe a strong positive correlation between biophysical destabilization and phenotypic deleteriousness, demonstrating that the deleteriousness of quasi-null mutations spans a continuous spectrum, with nonsense mutations at the extreme (highly deleterious) end. Our findings substantiate the disparity in phenotypic severity between missense and nonsense mutations and suggest that mutation-induced protein destabilization is indicative of the phenotypic outcomes of missense mutations. Our analyses of node-removal mutations allow for the potential identification of proteins whose removal or destabilization lead to harmful phenotypes, enabling the development of targeted therapeutic approaches, and enhancing comprehension of the intricate mechanisms governing genotype-to-phenotype relationships in clinically relevant diseases.

Transaminases have important applications in the synthesis of drug intermediates such as chiral amines. However, natural transaminases exhibit suboptimal thermal stability, limiting their further applications. Building upon an Rhodobacter sp.-derived (R)-selective transaminase (RbTA), we report a dual-region coupling engineering approach to improve thermostability of RbTA by strengthening the core hydrogen-bond networks and rigidifying the flexible surface loop. Through single strategy, we identified 4 thermostability improved single mutations, among which I249Q demonstrated the most substantial improvement, achieving a 18-fold increase in half-life (t1/240) and a 11.2 ℃ increase in T5010. Then in strategic coupling, the synergistic effect of dual-region modification was observed in both thermal stability and activity enhancement, as mutant with the best high-temperature catalytic performance, R136P/F228Y, had its T5010 improved by 7.1℃ and exhibited a 4.2-fold increase in kcat/Km towards (R)-3-amino-1-butanol. Finally, R136P/F228Y achieved a 20.5 % improvement in conversion over WT in an analytical-scale synthesis in 72 h at a 5 ℃ elevated catalytic temperature. Molecular dynamics simulations demonstrated that the synergy of the formation of new hydrogen bonds and decrease in flexibility accounted for the thermostability improvements. This study provides guidance for enhancing thermostability of similar fold-type enzymes without impairing enzymatic activity in an efficient manner.

The NIMA-related kinase 1 (NEK1) gene belongs to the Never in Mitosis Gene A (NIMA) kinase family, a group whose members play essential roles in cell cycle regulation, specifically in cell division and ciliogenesis. Mutations in the NEK1 gene have been implicated in several diseases, including short-rib polydactyly syndrome (SRPS). SRPS is a bone growth disorder characterized by severe congenital anomalies. Here, we describe a family with a lethal form of SRPS due to a novel intronic variant in the NEK1 gene. Basing on whole-exome sequencing of fetuses with SRPS we identified a homozygous variant of the NEK1 gene at position c.3584-10T>A as the causative mutation. Bioinformatic methods and minigene splicing assays were then used to assess the harmfulness and functional impact of the variant. We found that the identified mutation leads to the synthesis of the NEK1 protein lacking 90C-terminal residues following the last coiled-coil region. Additional experiments were performed to identify proteins that interact with the C-terminal fragment of NEK1 absent in the mutated protein. We suggest that the interaction between the C-terminal fragment of NEK1 and the VDAC1 channel is essential for the VDAC1 phosphorylation, the absence of which is likely to affect ciliogenesis.

Spectrins are ubiquitous cytoskeleton proteins found in all metazoan cells. αII-spectrin, encoded by SPTAN1, is the pivotal protein responsible for organization of the axonal cytoskeleton. Monoallelic SPTAN1 mutations cause various inherited neurological diseases, including spastic paraplegia 91 (SPG91), a type of hereditary spastic paraplegia (HSP).

We reported two patients with SPG91 caused by the SPTAN1 mutation c.55 C > T (p.Arg19Trp), who presented with lower limb spasticity and polyneuropathy. An analysis of the patients reported in the literature in addition to the present patients revealed that SPTAN1 p.Arg19Trp was specific for an HSP phenotype, with 35% of the combined patients with sensory‒motor polyneuropathy and 30% with cerebellar ataxia. In computational simulations, this variant was predicted to perturb the stability of αII/β spectrin heterotetramerization but did not destabilize the tetramerization domain of αII-spectrin.

Our findings on genotype‒phenotype correlations and genetic effects on molecular characteristics may provide important insights into the exploration of αII-spectrin-related neurological diseases.

The COVID-19 pandemic has prompted an unprecedented global response. In particular, extraordinary efforts have been dedicated toward monitoring and predicting variant emergence due to its huge impact, particularly for vaccine escape. Broadly, we classify such methods into two categories: forward mutation prediction, where phenotypes are first observed and the responsible genotypes traced, and reverse mutation prediction, which starts with selected pathogen genetic profiles and characterizes their associated phenotypes. Reverse mutation prediction strategies have advantages in being able to sample a more complete evolutionary space since sequences that do not yet exist can be sampled. The rapid improvement in the maturity and scale of reverse mutation prediction strategies, such as deep mutational scanning, has led to significant amounts of data for machine learning, with concomitant improvement in the prediction results from computational tools. Such integrated prediction approaches are generalizable and offer significant opportunities for anticipating viral evolution and for pandemic preparedness.

Protein-protein interactions are crucial for cellular regulation, antigen-antibody interactions, and other vital processes within living organisms. However, mutations in amino acid residues have the potential to induce changes in protein-protein binding affinity (ΔΔG), which may contribute to the onset and progression of disease. Existing methods for predicting ΔΔG use either protein sequence information or structural data. Furthermore, some methods are only applicable to single-point mutation cases. To address these limitations, we introduce a ΔΔG predictor that can handle complex scenarios involving multipoint mutations. In this investigation, a dual-channel deep learning model three-dimensional (3D)-ΔΔG is introduced, which is designed to predict ΔΔG by combining mutation information from side chain sequences and 3D structures. The proposed model employs a pre-trained protein language model to encode the side-chain amino acid sequence. A graph attention network is deployed to handle the graph representation of proteins simultaneously. Finally, a dual-channel processing module is implemented to facilitate depth fusion and extraction of both sequence and structural features. The model effectively captures the intricate alterations occurring pre- and post-protein mutation by integrating both sequence and 3D structural information. Results on the single-point mutation data set demonstrate a substantial improvement compared to state-of-the-art models. More significantly, 3D-ΔΔG exhibits superior performance when evaluated on the mixed mutation data sets, SKEMPIv1 and SKEMPIv2. The high level of agreement between the computationally predicted ΔΔG values and the experimentally determined values illustrates the potential of the 3D-ΔΔG model as an effective pre-screening tool in protein design and engineering.

The evolution of enzymes through multisite combinations is often constrained by the dynamic correlations among residues, resulting in the challenge of achieving additive effects when combining single-site positive mutants. Nevertheless, the incorporation of distal-site residue mutants offers a promising avenue to unlock synergistic interactions, particularly in enzymes with channels that mediate substrate and product transport. While combinatorial mutagenesis typically requires exhaustive screening of extensive mutant libraries to identify superior variants. In this study, a multisite combination strategy using dynamic cross-correlation matrices (DCCM) iterative analysis, coupled with location orthogonal filtration (DCCM/OL) was established based on single-site mutants enhancing enzyme performance by engineering distal residues using the "Structure, SCANEER, and Sequence (3S)" approach. Thirteen beneficial single mutants of oleate hydratase from Staphylococcus aureus (SaOhy), which catalyzes the hydration of linoleic acid, were targeted using the "3S" approach. By employing the DCCM/OL strategy, the number of candidates in the combination mutation library for experimental screening were reduced from 60 to 5. And a triple-site mutant, SaOhy_L151V/I411L/V135A, was ultimately identified, which increased the catalytic efficiencies with linoleic acid and oleic acid by a 4- and 2.3-fold, respectively. The philosophy of multisite mutation based on distal residues using the DCCM/OL strategy effectively achieves an amplification of enzymatic activity with distinct substrates simultaneously. This study offers a promising strategy for the construction of a mutant smart library and multisite combinations to efficiently increase enzymatic performance.

Enzymes are indispensable for biological processes and diverse applications across industries. While top-down modification strategies, such as directed evolution, have achieved remarkable success in optimizing existing enzymes, bottom-up de novo enzyme design has emerged as a transformative approach for engineering novel enzymes with customized catalytic functions, independent of natural templates. Recent advancements in artificial intelligence (AI) and computational power have significantly accelerated this field, enabling breakthroughs in enzyme engineering. These technologies facilitate the rapid generation of enzyme structures and amino acid sequences optimized for specific functions, thereby enhancing design efficiency. They also support functional validation and activity optimization, improving the catalytic performance, stability, and robustness of de novo designed enzymes. This review highlights recent advancements in AI-driven de novo enzyme design, discusses strategies for validation and optimization, and examines the challenges and future prospects of integrating these technologies into enzyme development.

Understanding the effects of missense mutations or single amino acid variants (SAVs) on protein function is crucial for elucidating the molecular basis of diseases/disorders and designing rational therapies. We introduce here Rhapsody-2, a machine learning tool for discriminating pathogenic and neutral SAVs, significantly expanding on a precursor limited by the availability of structural data. With the advent of AlphaFold2 as a powerful tool for structure prediction, Rhapsody-2 is trained on a significantly expanded dataset of 117,525 SAVs corresponding to 12,094 human proteins reported in the ClinVar database. Adopting a broad set of descriptors composed of sequence evolutionary, structural, dynamic, and energetics features in the training algorithm, Rhapsody-2 achieved an AUROC of 0.94 in 10-fold cross-validation when all SAVs of a particular test protein (mutant) were excluded from the training set. Benchmarking against a variety of testing datasets demonstrated the high performance of Rhapsody-2. While sequence evolutionary descriptors play a dominant role in pathogenicity prediction, those based on structural dynamics provide a mechanistic interpretation. Notably, residues involved in allosteric communication and those distinguished by pronounced fluctuations in the high-frequency modes of motion or subject to spatial constraints in soft modes usually give rise to pathogenicity when mutated. Overall, Rhapsody-2 provides an efficient and transparent tool for accurately predicting the pathogenicity of SAVs and unraveling the mechanistic basis of the observed behavior, thus advancing our understanding of genotype-to-phenotype relations.

Homologous recombination is a vital biological process for DNA repair, genomic stability, and genetic diversity, driven by the RecA/Rad51 recombinase family. However, as a T4 bacteriophage recombinase homologous to RecA/Rad51, UvsX has limited in vitro performance during recombinase polymerase amplification (RPA) due to ATP utilization and DNA affinity. In this study, UvsX was rationally engineered to enhance these properties through homology modeling, virtual saturation mutations, and consensus mutation strategies. Targeted mutagenesis produced UvsX variants (E198N, E198R, E198K, and K35G) with a 16 ± 4% to 39 ± 6% improvement in RPA activity, while the double mutant K35G/E198R showed an increase of up to 43 ± 4%. Structural analysis revealed that the K35G/E198R mutation enlarged ATP-binding pockets and increased the positive surface potential of DNA-binding sites, resulting in a 12 ± 4% improvement in ATP utilization and more ADP and less AMP generated, a 10 ± 2% enhancement in DNA interaction compared to the wild-type, and better inhibitor tolerance. These findings establish a foundation for the rational optimization of recombinases in nucleic acid amplification and promote their potential for industrial RPA applications.

Membrane contact sites (MCS) between the plasma membrane (PM) and endoplasmic reticulum (ER) regulate Ca2+ influx. However, the mechanisms by which cells modulate ER-PM MCS density are not understood, and the role of Ca2+, if any, in regulating these is unknown. We report that in Drosophila photoreceptors, MCS density is regulated by the Ca2+ channels, TRP and TRPL. Regulation of MCS density by Ca2+ is mediated by Drosophila extended synaptotagmin (dEsyt), a protein localized to ER-PM MCS and previously shown to regulate MCS density. We find that the Ca2+-binding activity of dEsyt is required for its function in vivo. dEsytCaBM, a Ca2+ non-binding mutant of dEsyt is unable to modulate MCS structure. Further, reconstitution of dEsyt null photoreceptors with dEsytCaBM is unable to rescue ER-PM MCS density and other key phenotypes. Thus, our data supports a role for Ca2+ binding to dEsyt in regulating ER-PM MCS density in photoreceptors thus tuning signal transduction during light-activated Ca2+ influx.

Epidermolysis bullosa simplex (EBS) refers to a heterogeneous group of inherited skin disorders characterized by blister formation within the basal cell layer. The disease is characterized by marked variations in phenotype severity, suggesting co-inheritance of genetic modifiers. We identified three deleterious variants in HMCN1 that co-segregated with a more severe phenotype in a group of 20 individuals with EBS caused by mutations in KRT14, encoding keratin 14 (K14). HMCN1 codes for hemicentin-1. Protein modeling, molecular dynamics simulations, and functional experiments showed that all three HMCN1 variants disrupt protein stability. Hemicentin-1 was found to be expressed in human skin above the BMZ. Using yeast-2-hybrid, co-immunoprecipitation, and proximity ligation assays, we found that hemicentin-1 binds K14. Three-dimensional skin equivalents grown from hemicentin-1-deficient cells were found to spontaneously develop subepidermal blisters, and HMCN1 downregulation was found to reduce keratin intermediate filament formation. In conclusion, hemicentin-1 binds K14 and contributes to BMZ stability, which explains the fact that deleterious HMCN1 variants co-segregate with a more severe phenotype in KRT14-associated EBS.

The heterologous synthesis of a nitrogen-fixing system in a non-diazotrophic organism is a long-sought-after goal because of the crucial importance of nitrogenase for agronomy, energy, and the environment. Here, we report the heterologous synthesis of a two-component nitrogenase analog from Azotobacter vinelandii, which consists of the reductase component (NifH) and the cofactor maturase (NifEN), in Escherichia coli. Metal, electron paramagnetic resonance, and activity analyses verify the cluster composition and functional competence of the heterologously expressed NifH and NifEN. Nuclear magnetic resonance, nanoscale secondary ion mass spectrometry, and growth experiments further illustrate the ability of the NifH/NifEN system to reduce N2 and incorporate the reduced N into the cellular mass. These results establish NifEN/NifH as a simplified nitrogenase analog that could be optimized and engineered to facilitate transgenic expression and biotechnological adaptations of this important metalloenzyme.

Classification of missense variants is challenging. Lacking compelling clinical and/or functional data, ACMG/AMP lines of evidence are restricted to PM2 (rarity code applied at supporting level) and PP3/BP4 (computational evidence based mostly on multiple-sequence-alignment conservation tools). Currently, the ClinGen ENIGMA BRCA1/2 Variant Curation Expert Panel uses BayesDel to apply PP3/BP4 to missense variants located in the BRCA1 RING/BRCT domains. The ACMG/AMP framework does not refer explicitly to protein structure as a putative source of pathogenic/benign evidence. Here, we tested the value of incorporating structure-based evidence such as relative solvent accessibility (RSA), folding stability (ΔΔG), and/or AlphaMissense pathogenicity to the classification of BRCA1 missense variants. We used MAVE functional scores as proxies for pathogenicity/benignity. We computed RSA and FoldX5.0 ΔΔG predictions using as alternative input templates for either PDB files or AlphaFold2 models, and we retrieved pre-computed AlphaMissense and BayesDel scores. We calculated likelihood ratios toward pathogenicity/benignity provided by the tools (individually or combined). We performed a clinical validation of major findings using the large-scale BRIDGES case-control dataset. AlphaMissense outperforms ΔΔG and BayesDel, providing similar PP3/BP4 evidence strengths with lower rate of variants in the uninformative score range. AlphaMissense combined with ΔΔG increases evidence strength granularity. AlphaFold2 models perform well as input templates for ΔΔG predictions. Regardless of the tool, BP4 (but not PP3) is highly dependent on RSA, with benignity evidence provided only to variants targeting buried or partially buried residues (RSA ≤ 60%). Stratification by functional domain did not reveal major differences. In brief, structure-based analysis improves PP3/BP4 assessment, uncovering a relevant role for RSA.

Deep learning methods have played an increasingly pivotal role in advancing side-chain packing and mutation effect prediction (ΔΔG) for protein complexes. Although these two tasks are inherently closely related, they are typically treated separately in practice. Furthermore, the lack of effective post-processing in most approaches results in sub-optimal refinement of generated conformations, limiting the plausibility of the predicted conformations. In this study, we introduce an integrated framework, PackPPI, which employs a diffusion model and a proximal optimization algorithm to improve side-chain prediction for protein complexes while using learned representations to predict ΔΔG. The results demonstrate that PackPPI achieved the lowest atom RMSD (0.9822) on the CASP15 dataset. The proximal optimization algorithm effectively reduces spatial clashes between side-chain atoms while maintaining a low-energy landscape. Furthermore, PackPPI achieves state-of-the-art performance in predicting binding affinity changes induced by multi-point mutations on the SKEMPI v2.0 dataset. These findings underscore the potential of PackPPI as a robust and versatile computational tool for protein design and engineering. The implementation of PackPPI is available at https://github.com/Jackz915/PackPPI.

β-Glucanases, widely applied in grain processing, are commonly restricted for efficient industrial application due to the limited thermostability. In this study, a 1,3-1,4-β-glucanase (BisGlu16B_ΔC) was optimized for thermostability through a computer-aided design of energy optimization. Three variants (T40K, Q53L, and S311Y) were selected and generated a combined mutant T40K/Q53L/S311Y (M3). Comparing with the WT, M3 exhibited better thermostability (with t1/2 at 60 °C extend by 126 min), higher specific activity (1.24 folds; 69,700 vs. 56,200 U/mg), higher catalytic effciency (1.18 folds; 14,100 vs. 11,900 mL‧s-1‧mg-1), and improved protease resistance. For mechanism, more hydrogen bonds, salt bridges, and rigid secondary components in M3 led to an enhanced overall rigidity, boosting the thermostability. While enhanced long-range negative interactions affected some key residues in the catalytic channels, improving the catalytic efficiency. For application, M3 showed superiority with higher dry matter digestibility (1.49 folds; 80.3 % vs. 53.9 %) in simulated gastrointestinal system, together with more reduction of filtration time (1.55 folds; 22.2 % vs. 14.3 %) and viscosity (2.37 folds; 10.2 % vs. 4.3 %) during malting, comparing with the WT. Furthermore, the strongest synergistic effects were found between xylanase and M3, among all β-glucanases tested. All results verified M3 as an efficient β-glucanase for grain processing industry.

Engineering protein stability is a critical challenge in biotechnology. Here, we used massively parallel deep mutational scanning (DMS) to comprehensively explore the mutational stability landscape of human myoglobin (hMb) and identify key mutations that enhance stability. Our DMS approach involved screening over 10,000 hMb variants by yeast surface display, single-cell sorting, and high-throughput DNA sequencing. We show how surface display levels serve as a proxy for thermostability of soluble hMb variants and report strong correlations between DMS-derived display levels and top-performing machine learning stability prediction algorithms. This approach led to the discovery of a variant with a de novo disulfide bond between residues R32C and C111, which increased thermostability by >12°C compared with wild-type hMb. By combining single stabilizing mutations with R32C, we engineered combinatorial variants that exhibited predominantly additive effects on stability with minimal epistasis. The most stable combinatorial variant exhibited a denaturation temperature exceeding 89°C, representing a >17°C improvement over wild-type hMb. Our findings demonstrate the capabilities in DMS-assisted combinatorial protein engineering to guide the discovery of thermostable variants and highlight the potential of massively parallel mutational analysis for the development of proteins for industrial and biomedical applications.

Predicting protein stability changes upon single-point mutations is crucial in computational biology, with applications in drug design, enzyme engineering, and understanding disease mechanisms. While deep-learning approaches have emerged, many remain inaccessible for routine use. In contrast, potential-like methods, including deep-learning-based ones, are faster, user-friendly, and effective in estimating stability changes. However, most of them approximate Gibbs free-energy differences without accounting for the free-energy changes of the unfolded state, violating mass balance and potentially reducing accuracy. Here, we show that incorporating mass balance as a first approximation of the unfolded state significantly improves potential-like methods. While many machine-learning models implicitly or explicitly use mass balance, our findings suggest that a more accurate unfolded-state representation could further enhance stability change predictions.

Structural biology offers a powerful lens through which to assess genetic variants by providing insights into their impact on clinically relevant protein structure and function. Due to the availability of new, user-friendly, web-based tools, structural analyses by wider audiences have become more mainstream. These new tools, including AlphaMissense and AlphaFold, have recently been in the limelight due to their initial success and projected future promise; however, the intricacies and limitations of using these tools still need to be disseminated to the more general audience that is likely to use them in variant analysis. Here, we expound on frameworks applying structural biology to variant interpretation by examining two accompanying articles. To this end, we explore the nuances of choosing the correct protein model, compare and contrast various structural approaches, and highlight both the advantages and limitations of employing structural biology in variant interpretation. Using two articles published in this issue of The American Journal of Human Genetics as a baseline, we focus on case studies in TP53 and BRCA1 to illuminate gene-specific differences in the applications of structural information, which illustrate the complexities inherent in this field. Additionally, we discuss the implications of recent advancements, such as AlphaFold, and provide practical guidance for researchers navigating variant interpretation using structural biology.

The clinical classification of germline missense variants and single-amino-acid deletions is challenging. The BayesDel and Align-GVGD bioinformatic prediction tools currently used for ClinGen TP53 variant curation expert panel (VCEP) classification do not directly capture changes in protein folding stability, measured using computed destabilization energies (ΔΔG scores). The AlphaMissense tool recently developed by Google DeepMind to predict pathogenicity for all human proteome missense variants is trained in part using AlphaFold2 architecture. Our study investigated whether protein folding stability and/or AlphaMissense scores could improve impact prediction for p53 missense and single-amino-acid deletion variants. ΔΔG scores were calculated for missense variants using FoldX and for single-amino-acid deletions using an AlphaFold2/RosettaRelax protocol. Residue surface exposure was categorized using relative solvent accessibility (RSA) measures. The predictive values of ΔΔG scores, AlphaMissense, BayesDel, and Align-GVGD were examined using Boruta and binary logistic regression based on functionally defined reference sets. The likelihood ratio (LR) toward pathogenicity was estimated and used to refine optimal categories for predicting variant pathogenicity for different RSA values. We showed that current VCEP predictive approaches for missense variants were improved by integrating ΔΔG scores ≥2.5 kcal/mol for partially buried and buried residues, but better performance was achieved using AlphaMissense with ΔΔG and RSA. For deletion variants, ΔΔG scores ≥4.8 Rosetta energy unit (REU) in buried residues outperformed currently used predictive approaches. Future TP53 VCEP specifications for p53 missense impact prediction may consider AlphaMissense, ΔΔG score, and RSA combined for substitution variants and ΔΔG score alone for deletion variants.

In the field of industrial biocatalysis, the rapid advancement of enzyme functional evolution necessitates new theories and computational methods to achieve target functions with fewer iterations. This study identified key residues affecting enzyme stability by constructing the protein contact network (PCN) of Lipase A. Comparing the PCNs of the wild-type (WT) and the 6B variant revealed that changes in residue interactions and node properties (e.g., degree and betweenness centrality (BC)) positively impacted stability. Using thresholds for degree and BC, 25 candidate sites were screened, and 11 out of 18 single-point mutation designs improved thermal stability. Mutations were divided into three groups (M1, M2, M3) based on network communities and contributions, followed by iterative combinations. M1, containing five mutations distributed across four communities, increased the melting temperature (Tm) by 14.61 °C, close to the predicted 13.97 °C, demonstrating a linear additive effect. In M2, three new mutations resulted in a non-linear additive effect, with a ΔTm of 17.58 °C (Expected ΔTm = 18.93 °C). In contrast, the three new mutations in M3 destabilized the enzyme (Observed ΔTm = 15.94 °C vs Expected ΔTm = 19.92 °C). Molecular dynamics simulations showed that polar edge nodes enhanced network connectivity, while proline mutations rigidified flexible regions, improving stability. Conversely, M3 mutations disrupted α-helix stability by increasing the dihedral angle fluctuations of residue Y161, might to a stability-activity trade-off. The PCN provides valuable insights for developing efficient and precise design strategies.

Missense mutations in oncogenic proteins that are concurrently associated with neurodevelopmental disorders have garnered significant attention. Phosphatase and tensin homologue (PTEN) serves as a paradigmatic model for mapping its mutational landscape and identifying genotypic predictors of distinct phenotypic outcomes, including cancer and autism spectrum disorder (ASD). Despite extensive research into the genotype-phenotype correlations of PTEN mutations, the mechanisms underlying the dual association of specific PTEN mutations with both cancer and ASD (PTEN-cancer/ASD mutations) remain elusive. This study introduces an integrative approach that combines machine learning (ML) with structural dynamics to elucidate the molecular effects of PTEN-cancer/ASD mutations. Analysis of biophysical and network-biology-based signatures reveals a complex energetic and functional landscape. Subsequently, an ML model and corresponding integrated score were developed to classify and predict PTEN-cancer/ASD mutations, underscoring the significance of protein dynamics in predicting cellular phenotypes. Further molecular dynamics simulations demonstrated that PTEN-cancer/ASD mutations induce dynamic alterations characterized by open conformational changes restricted to the P loop and coupled with interdomain allosteric regulation. This research aims to enhance the genotypic and phenotypic understanding of PTEN-cancer/ASD mutations through an interpretable ML model integrated with structural dynamics analysis. By identifying shared mechanisms between cancer and ASD, the findings pave the way for the development of novel therapeutic strategies.

Congenital hypofibrinogenemia presents not only with bleeding, but also paradoxically with thrombosis. This heterogeneity of clinical phenotype complicates both diagnosis and management. The thrombotic phenotype is thought to arise from alterations in fibrin structure and stability, leading to abnormal clot formation and an increased risk of thrombosis. Coagulation assays, gene analysis, and protein modeling were utilized to elucidate the pathogenic variant. We highlight the pathophysiology of the novel missense variant in the FGG gene (c.823G/A, p.Glu275Lys), which causes mild hypofibrinogenemia and clinically manifests as an ischemic stroke. Protein modeling displays that the amino-acid substitution of glutamine with lysine at position 275 in mentioned missense variant causes local changes in the fibrinogen structure. The structural changes are mainly minor surface alterations and changes in physicochemical properties, which could potentially affect the recruitment of other proteins or lead to abnormal fibrin polymerization. This study provides novel insights into the pathophysiological mechanism, emphasizing the importance of molecular and structural analyses in understanding and managing atypical presentations of fibrinogen disorders.

Advanced genetic strategies have transformed our understanding of the genetic basis and diagnosis of many phenotypes, including rare diseases. However, missense variants (MVs) are frequently identified and often classified as variants of uncertain significance (VUS). Although changes in protein free energy (ΔΔG) were recently proposed as a tool for VUS classification, no objective cut-offs exist to distinguish between benign and pathogenic variants.

We utilized the computational tool mCSM to calculate ΔΔG and predict the impact of MVs on protein stability. Specifically, we systematically analyzed the ΔΔG of MVs in IFT140 to identify those potentially pathogenic and associated with Mainzer-Saldino syndrome (MSS). To this end, we evaluated ΔΔG in IFT140 MVs sourced from ClinVar, gnomAD, and MSS patients, aiming to resolve the diagnosis of MSS in a child with a novel homozygous IFT140 variant, initially reported as a VUS.

IFT140 MVs from MSS patients showed lower ΔΔG values than those reported in gnomAD individuals (-1.389 vs. -0.681 kcal/mol; p = 0.0031). A ROC curve demonstrated strong discriminative ability (AUC = 0.8488; p = 0.0002), and a ΔΔG cut-off of -1.3 kcal/mol achieving 50% sensibility and 90% specificity. The analysis of ClinVar IFT140 variants classified as VUS, showed that 75/323 (23%) presented ΔΔG values below the cut-off. In the child clinically suspicious of MSS, this cut-off allowed the reclassification of the VUS (IFT140:p.W80C; ΔΔG = -1.745 kcal/mol) as likely pathogenic, which confirmed the diagnosis molecularly.

Our findings demonstrate that ΔΔG analysis can effectively distinguish potentially pathogenic variants in IFT140, enabling confirmation of MSS. The established cut-off of -1.3 kcal/mol showed strong discriminative power, aiding in the reclassification of VUS identified in IFT140. This approach highlights the utility of protein stability predictions in resolving diagnostic uncertainty in rare diseases.

Aspergillus fumigatus is a common saprophytic filamentous fungus that plays a crucial role in nutrient cycling but can become an opportunistic pathogen, posing a significant threat to immunocompromised individuals by causing invasive aspergillosis. A key feature of A. fumigatus is the presence of hydrophobins-small amphipathic proteins that form a protective rodlet layer on conidial surfaces, facilitating biofilm formation and immune evasion. This rodlet structure, stabilized by cross-linking disulfide bonds, provides resistance to desiccation, oxidative stress, and immune defenses, making these cross-links a compelling target for study. In this work, we employ all-atom simulations, incorporating quantum mechanics/molecular mechanics (QM/MM) calculations, to evaluate the energy and conformational effects of cross-linking disulfide bonds (CL1, CL2, CL3, and CL4) in the rodlet assembly. By integrating QM/MM approaches, we achieve a detailed representation of the electronic and structural properties of these bonds within the complex rodlet layer, gaining deeper insights into their essential role in maintaining the stability and integrity of the RodA hydrophobin protein from A. fumigatus conidial surface. We identify a group of ten residues that influence directly in the cross-linking, with Gln23 and Lys17 emerging as key candidates for experimental mutation to control rodlet assembly. Our findings shed light on the molecular mechanisms underlying rodlet formation and highlight potential targets for disrupting this protective layer, offering promising avenues for antifungal strategies.

Developmental and epileptic encephalopathy type 50 (DEE50) in humans is a severe early-onset neurometabolic disorder caused by biallelic loss-of-function variants in the CAD gene encoding a key multi-enzymatic protein for de novo pyrimidine nucleotide synthesis. Untreated, the condition is often fatal, but patients respond to uridine supplementation, which fuels nucleotide synthesis through CAD-independent salvage pathways. Here, we report a novel variant in the feline CAD gene in a 4-month-old Bengal kitten with intractable seizures and abnormal behavior. The variant, XP_011279586.1:p.(Ser2015Asn), was predicted to affect the oligomerization of the C-terminal aspartate transcarbamylase (ATCase) domain of CAD. Genotyping of 110 unaffected Bengal cats revealed four additional carriers of the mutant allele, confirming its presence in the breed. In a CAD-knockout human cell line dependent on uridine, the recombinant expression of human wildtype CAD, but not of the Asn2015 mutant, restored cell growth without uridine, demonstrating that the p.Ser2015Asn variant disrupts CAD function and is pathogenic. This study facilitates genetic testing of carriers and affected cats and suggests that uridine supplementation could be a potential treatment. Furthermore, CAD-deficient Bengal cats might serve as a valuable spontaneous large animal model to further investigate the pathogenic mechanisms of this rare epileptic encephalopathy in humans.

Proteins play a critical role in biology and biopharma due to their specificity and minimal side effects. Predicting the effects of mutations on protein stability is vital but experimentally challenging. Deep learning offers an efficient solution to this problem. In the present work, we introduced ProstaNet, a deep learning framework that predicts stability changes resulting from single- and multiple-point mutations using geometric vector perceptrons-graph neural network for 3-dimensional feature processing. For training ProstaNet, we meticulously crafted ProstaDB, a comprehensive and pristine thermodynamics repository, including 3,784 single-point mutations and 1,642 multiple-point mutations. We also created thermodynamic looping for enlarging the limited data size of multiple-point mutation and applied an innovative clustering method to generate a standard testing set of multiple-point mutation. Besides, we identified residue scoring as the most important encoding method in protein properties prediction. With these innovations, ProstaNet accurately predicts thermostability changes for both single-point and multiple-point mutations without showing any bias. ProstaNet achieves an accuracy of 0.75, outperforming existing methods for single-point mutation prediction, including ThermoMPNN (0.63), PoPMuSiCsym (0.66), MUPRO (0.52), and FoldX (0.71). ProstaNet also achieves a 1.3-fold increase in accuracy compared to FoldX for multiple-point mutation predictions. Validated by experiment, 4 out of 5 single-point mutation predictions (80%) and all multiple-point mutation predictions (100%) for HuJ3 mutants were accurate, demonstrating the potential benefits of ProstaNet for protein engineering and drug development.

Accurately predicting the effect of mutations on protein-protein interactions (PPIs) is essential for understanding the protein structure and function, as well as providing insights into disease-causing mechanisms. Many recent popular approaches based on the three-dimensional structure of proteins have been proposed to predict the changes in binding affinity caused by mutations, i.e. ΔΔG. However, how to effectively use the structural information to comprehensively exploit complex interactions within proteins and integrate multisource features remains a significant challenge. In this study, we propose SFM-Net, a powerful deep learning model constructed with GNN-based multiway feature extractors and a new context-aware selective fusion module that jointly leverages the sequence, structural, and evolutionary information. Such design enables SFM-Net to effectively and selectively use features from different sources to facilitate binding affinity change prediction. Benchmarking experiments and targeted ablation studies illustrate the effectiveness and robustness of our method for improving the binding affinity change prediction.

Mycobacterium abscessus is one of the leading causes of pulmonary infections caused by non-tuberculous mycobacteria. The ability of M. abscessus to establish a chronic infection in the lung relies on a series of adaptive mutations impacting, in part, global regulators and cell envelope biosynthetic enzymes. One of the genes under strong evolutionary pressure during host adaptation is ubiA, which participates in the elaboration of the arabinan domains of two major cell envelope polysaccharides: arabinogalactan (AG) and lipoarabinomannan (LAM). We here show that patient-derived UbiA mutations not only cause alterations in the AG, LAM, and mycolic acid contents of M. abscessus but also tend to render the bacterium more prone to forming biofilms while evading uptake by innate immune cells and enhancing their pro-inflammatory properties. The fact that the effects of UbiA mutations on the physiology and pathogenicity of M. abscessus were impacted by the rough or smooth morphotype of the strain suggests that the timing of their selection relative to morphotype switching may be key to their ability to promote chronic persistence in the host.IMPORTANCEMultidrug-resistant pulmonary infections caused by Mycobacterium abscessus and subspecies are increasing in the U.S.A. and globally. Little is known of the mechanisms of pathogenicity of these microorganisms. We have identified single-nucleotide polymorphisms (SNPs) in a gene involved in the biosynthesis of two major cell envelope polysaccharides, arabinogalactan and lipoarabinomannan, in lung-adapted isolates from 13 patients. Introduction of these individual SNPs in a reference M. abscessus strain allowed us to study their impact on the physiology of the bacterium and its interactions with immune cells. The significance of our work is in identifying some of the mechanisms used by M. abscessus to colonize and persist in the human lung, which will facilitate the early detection of potentially more virulent clinical isolates and lead to new therapeutic strategies. Our findings may further have broader biomedical impacts, as the ubiA gene is conserved in other tuberculous and non-tuberculous mycobacterial pathogens.

Evaluating the mutation impact on protein stability (ΔΔG) is essential in the study of protein engineering and understanding molecular mechanisms of disease-associated mutations. Here, we propose a novel deep learning framework, PILOT, for improved prediction of ΔΔG using a Siamese network with hybrid attention mechanism. The PILOT framework leverages multiple attention modules to effectively extract representations for amino acids, atoms, and protein sequences, respectively. This approach significantly ensures the deep fusion of structural information at both residue and atom levels, the seamless integration of structural and sequence representations, and the effective capture of both long-range and short-range dependencies among amino acids. Our extensive evaluations demonstrate that PILOT greatly outperforms other state-of-the-art methods. We also showcase that PILOT identifies exceptional patterns for different mutation types. Moreover, we illustrate the clinical applicability of PILOT in highlighting pathogenic variants from benign variants and VUS (variants of uncertain significance), and distinguishing de novo mutations in disease cases and controls. In summary, PILOT presents a robust deep learning tool that could offer significant insights into drug design, medical applications, and protein engineering studies.

Understanding the folding and stability of transmembrane β-barrel proteins (TMBs) provides insights into their structural integrity, functional mechanisms, and implications for disease states. In this work, we have characterized the important features that influence the folding and stability of TMBs. Our results showed that lipid accessible surface area and transition energy are important for understanding the stability of TMBs. Further, this information was utilized to develop a linear regression-based method for predicting the stability of TMBs. Our method achieved a correlation and mean absolute error (MAE) of 0.96 and 0.94 kcal/mol on the jack-knife test. Moreover, we compared the stability of TMBs with globular all-β proteins and observed that long-range interactions and energetic properties are crucial for maintaining the stability of both β-barrel membrane and all-β globular proteins. On the other hand, side-chain - side-chain hydrogen bonds and lipid accessible surface area are specific to membrane proteins. These features are critical for membrane proteins because they influence a protein to embed within the membrane environment. Further, we have developed a web server, TMB Stab-pred for predicting the stability of TMBs, and it is accessible at https://web.iitm.ac.in/bioinfo2/TMBB/index.html.

Identifying interactions between candidate antibodies and target antigens is a key step in developing effective human therapeutics. The antigen-antibody interaction (AAI) occurs at the structural level, but the limited structure data poses a significant challenge. However, recent studies revealed that structural information can be learned from the vast amount of sequence data, indicating that the interaction prediction can benefit from the abundance of antigen and antibody sequences. In this study, DeepInterAware (deep interaction interface-aware network) is proposed, a framework dynamically incorporating interaction interface information directly learned from sequence data, along with the inherent specificity information of the sequences. Experimental results in interaction prediction demonstrate that DeepInterAware outperforms existing methods and exhibits promising inductive capabilities for predicting interactions involving unseen antigens or antibodies, and transfer capabilities for similar tasks. More notably, DeepInterAware has unique advantages that existing methods lack. First, DeepInterAware can dive into the underlying mechanisms of AAIs, offering the ability to identify potential binding sites. Second, it is proficient in detecting mutations within antigens or antibodies, and can be extended for precise predictions of the binding free energy changes upon mutations. The HER2-targeting antibody screening experiment further underscores DeepInterAware's exceptional capability in identifying binding antibodies for target antigens, establishing it as an important tool for antibody screening.

Missense variants can affect the severity of disease, choice of treatment, and treatment outcomes. While the number of known variants has been increasing at a rapid pace, available evidence of their clinical effect has been lagging behind, constituting a challenge for clinicians and researchers. Multiplexed assays of variant effects (MAVEs) are important to close the gap; nonetheless, computational predictions of pathogenicity are still often the only available data for scoring variants. Such methods are not designed to provide a mechanistic explanation for the effect of amino acid substitutions. To this purpose, we propose structure-based frameworks as ensemble methodologies, with each method tailored to predict a different aspect among those exerted by amino acid substitutions to link predicted pathogenicity to mechanistic indicators. We review available frameworks, as well as advancements in underlying structure-based methods that predict variant effects on several protein features, such as protein stability, biomolecular interactions, allostery, post-translational modifications, and more.

High mutation and replication rates of HIV-1 result in the continuous generation of variants, allowing it to adapt to changing host environments. Mutations often have deleterious effects, but variants carrying them are rapidly purged. Surprisingly, a particular variant incapable of entering host cells was found to be rescued by host antibodies targeting HIV-1. Understanding the molecular mechanism of this rescue is important to develop and improve antibody-based therapies. To unravel the underlying mechanisms, we performed fully atomistic molecular dynamics simulations of the HIV-1 gp41 trimer responsible for viral entry into host cells, its entry-deficient variant, and its complex with the rescuing antibody. We find that the Q563R mutation, which the entry-deficient variant carries, prevents the native conformation of the gp41 6-helix bundle required for entry and stabilizes an alternative conformation instead. This is the consequence of substantial changes in the secondary structure and interactions between the domains of gp41. Binding of the antibody F240 to gp41 reverses these changes and re-establishes the native conformation, resulting in rescue. To test the generality of this mechanism, we performed simulations with the entry-deficient L565A variant and antibody 3D6. We find that 3D6 binding was able to reverse structural and interaction changes introduced by the mutation and restore the native gp41 conformation. Viral variants may not only escape antibodies but be aided by them in their survival, potentially compromising antibody-based therapies, including vaccination and passive immunization. Our simulation framework could serve as a tool to assess the likelihood of such resistance against specific antibodies.

The spike protein encoded by the SARS-CoV-2 has become one of the most studied macromolecules in recent years due to its central role in COVID-19 pathogenesis. The spike protein's receptor-binding domain (RBD) directly interacts with the host-encoded receptor protein, ACE2. This review critically examines computational insights into RBD's interaction with ACE2 and with therapeutic antibodies designed to interfere with this interaction. We begin by summarizing insights from early computational studies on pre-pandemic SARS-CoV-1 RBD interactions and how these early studies shaped the understanding of SARS-CoV-2. Next, we highlight key theoretical contributions that revealed the molecular mechanisms behind the binding affinity of SARS-CoV-2 RBD against ACE2, and the structural changes that have enhanced the infectivity of emerging variants. Special attention is given to the "RBD charge rule", a predictive framework for determining variant infectivity based on the electrostatic properties of the RBD. Towards applying the computational insights to therapy, we discuss a multiscale computational protocol for optimizing monoclonal antibodies to improve binding affinity across multiple spike protein variants, including representatives from the Omicron family. Finally, we explore how these insights can inform the development of future vaccines and therapeutic interventions for combating future coronavirus diseases.

The mutation of remote positions on enzyme scaffolds and how these residue changes can affect enzyme catalysis is still far from being fully understood. One paradigmatic example is the group of lysosomal storage disorders, where the enzyme activity of a lysosomal enzyme is abolished or severely reduced. In this work, we analyze molecular dynamics simulation conformational ensembles to unveil the molecular features controlling the deleterious effects of the 43 reported missense mutations in the human lysosomal α-mannosidase. Using residue descriptors for protein dynamics, their coupling with the active site, and their impact on protein stability, we have assigned the contribution of each of the missense mutations into protein stability, protein dynamics, and their connectivity with the active site. We demonstrate here that the use of conformational ensembles is a powerful approach not only to better understand missense mutations at the molecular level but also to revisit the missense mutations reported in lysosomal storage disorders in order to aid the treatment of these diseases.

Predicting the evolutionary patterns of emerging and endemic viruses is key for mitigating their spread. In particular, it is critical to rapidly identify mutations with the potential for immune escape or increased disease burden. Knowing which circulating mutations pose a concern can inform treatment or mitigation strategies such as alternative vaccines or targeted social distancing. In 2021, Hie B, Zhong ED, Berger B, Bryson B. 2021 Learning the language of viral evolution and escape. Science 371, 284-288. (doi:10.1126/science.abd7331) proposed that variants of concern can be identified using two quantities extracted from protein language models, grammaticality and semantic change. These quantities are defined by analogy to concepts from natural language processing. Grammaticality is intended to be a measure of whether a variant viral protein is viable, and semantic change is intended to be a measure of potential for immune escape. Here, we systematically test this hypothesis, taking advantage of several high-throughput datasets that have become available, and also comparing this model with several more recently published machine learning models. We find that grammaticality can be a measure of protein viability, though methods that are trained explicitly to predict mutational effects appear to be more effective. By contrast, we do not find compelling evidence that semantic change is a useful tool for identifying immune escape mutations.