Quantitative Missense Variant Effect Prediction Using Large-Scale Mutagenesis Data

AlzDiscovery: A computational tool to identify Alzheimer's disease-causing missense mutations using protein structure information

Alzheimer's disease (AD) is one of the most common forms of dementia and neurodegenerative diseases, characterized by the formation of neuritic plaques and neurofibrillary tangles. Many different proteins participate in this complicated pathogenic mechanism, and missense mutations can alter the folding and functions of these proteins, significantly increasing the risk of AD. However, many methods to identify AD-causing variants did not consider the effect of mutations from the perspective of a protein three-dimensional environment. Here, we present a machine learning-based analysis to classify the AD-causing mutations from their benign counterparts in 21 AD-related proteins leveraging both sequence- and structure-based features. Using computational tools to estimate the effect of mutations on protein stability, we first observed a bias of the pathogenic mutations with significant destabilizing effects on family AD-related proteins. Combining this insight, we built a generic predictive model, and improved the performance by tuning the sample weights in the training process. Our final model achieved the performance on area under the receiver operating characteristic curve up to 0.95 in the blind test and 0.70 in an independent clinical validation, outperforming all the state-of-the-art methods. Feature interpretation indicated that the hydrophobic environment and polar interaction contacts were crucial to the decision on pathogenic phenotypes of missense mutations. Finally, we presented a user-friendly web server, AlzDiscovery, for researchers to browse the predicted phenotypes of all possible missense mutations on these 21 AD-related proteins. Our study will be a valuable resource for AD screening and the development of personalized treatment.

Screening and modification of (+)-germacrene A synthase for the production of the anti-tumor drug (-)-β-elemene in engineered Saccharomyces cerevisiae

(-)-β-Elemene is a primary bioactive compound derived from Curcuma wenyujin and has been widely utilized as an anti-tumor agent for various types of cancer. Due to the inefficiency of plant extraction methods for β-elemene, significant efforts have been directed toward the heterogeneous biosynthesis of β-elemene using microbial cell factories. However, there has been less emphasis on the stereochemical configuration of germacrene A and its rearranged product, β-elemene. In this study, we constructed a yeast cell factory to produce (-)-β-elemene by optimizing the mevalonate pathway and screening for germacrene A synthases (GASs) from both plant and microbial sources. Notably, we discovered that the rearranged products of GASs exhibited different conformations, and only (+)-germacrene A produced by plant-derived GASs could rearrange to form (-)-β-elemene. Building on this discovery, we further investigated the catalytic mechanisms of GASs and developed an efficient catalytic gene module for generating (+)-germacrene A. Ultimately, the engineered yeast produced 1152 mg/L of (-)-β-elemene, marking the highest titer reported in yeast to date. Overall, this work highlights the differences in the stereoconformations of catalytic products between plant- and microbial-derived germacrene A synthases and establishes a foundation for the green and efficient production of β-elemene with a specific stereochemical configuration.

PETA: evaluating the impact of protein transfer learning with sub-word tokenization on downstream applications

Protein language models (PLMs) play a dominant role in protein representation learning. Most existing PLMs regard proteins as sequences of 20 natural amino acids. The problem with this representation method is that it simply divides the protein sequence into sequences of individual amino acids, ignoring the fact that certain residues often occur together. Therefore, it is inappropriate to view amino acids as isolated tokens. Instead, the PLMs should recognize the frequently occurring combinations of amino acids as a single token. In this study, we use the byte-pair-encoding algorithm and unigram to construct advanced residue vocabularies for protein sequence tokenization, and we have shown that PLMs pre-trained using these advanced vocabularies exhibit superior performance on downstream tasks when compared to those trained with simple vocabularies. Furthermore, we introduce PETA, a comprehensive benchmark for systematically evaluating PLMs. We find that vocabularies comprising 50 and 200 elements achieve optimal performance. Our code, model weights, and datasets are available at https://github.com/ginnm/ProteinPretraining . SCIENTIFIC CONTRIBUTION: This study introduces advanced protein sequence tokenization analysis, leveraging the byte-pair-encoding algorithm and unigram. By recognizing frequently occurring combinations of amino acids as single tokens, our proposed method enhances the performance of PLMs on downstream tasks. Additionally, we present PETA, a new comprehensive benchmark for the systematic evaluation of PLMs, demonstrating that vocabularies of 50 and 200 elements offer optimal performance.

QAFI: a novel method for quantitative estimation of missense variant impact using protein-specific predictors and ensemble learning

Next-generation sequencing (NGS) has revolutionized genetic diagnostics, yet its application in precision medicine remains incomplete, despite significant advances in computational tools for variant annotation. Many variants remain unannotated, and existing tools often fail to accurately predict the range of impacts that variants have on protein function. This limitation restricts their utility in relevant applications such as predicting disease severity and onset age. In response to these challenges, a new generation of computational models is emerging, aimed at producing quantitative predictions of genetic variant impacts. However, the field is still in its early stages, and several issues need to be addressed, including improved performance and better interpretability. This study introduces QAFI, a novel methodology that integrates protein-specific regression models within an ensemble learning framework, utilizing conservation-based and structure-related features derived from AlphaFold models. Our findings indicate that QAFI significantly enhances the accuracy of quantitative predictions across various proteins. The approach has been rigorously validated through its application in the CAGI6 contest, focusing on ARSA protein variants, and further tested on a comprehensive set of clinically labeled variants, demonstrating its generalizability and robust predictive power. The straightforward nature of our models may also contribute to better interpretability of the results.

Flattening the curve-How to get better results with small deep-mutational-scanning datasets

Proteins are used in various biotechnological applications, often requiring the optimization of protein properties by introducing specific amino-acid exchanges. Deep mutational scanning (DMS) is an effective high-throughput method for evaluating the effects of these exchanges on protein function. DMS data can then inform the training of a neural network to predict the impact of mutations. Most approaches use some representation of the protein sequence for training and prediction. As proteins are characterized by complex structures and intricate residue interaction networks, directly providing structural information as input reduces the need to learn these features from the data. We introduce a method for encoding protein structures as stacked 2D contact maps, which capture residue interactions, their evolutionary conservation, and mutation-induced interaction changes. Furthermore, we explored techniques to augment neural network training performance on smaller DMS datasets. To validate our approach, we trained three neural network architectures originally used for image analysis on three DMS datasets, and we compared their performances with networks trained solely on protein sequences. The results confirm the effectiveness of the protein structure encoding in machine learning efforts on DMS data. Using structural representations as direct input to the networks, along with data augmentation and pretraining, significantly reduced demands on training data size and improved prediction performance, especially on smaller datasets, while performance on large datasets was on par with state-of-the-art sequence convolutional neural networks. The methods presented here have the potential to provide the same workflow as DMS without the experimental and financial burden of testing thousands of mutants. Additionally, we present an open-source, user-friendly software tool to make these data analysis techniques accessible, particularly to biotechnology and protein engineering researchers who wish to apply them to their mutagenesis data.

Guidelines for releasing a variant effect predictor

Computational methods for assessing the likely impacts of mutations, known as variant effect predictors (VEPs), are widely used in the assessment and interpretation of human genetic variation, as well as in other applications like protein engineering. Many different VEPs have been released to date, and there is tremendous variability in their underlying algorithms and outputs, and in the ways in which the methodologies and predictions are shared. This leads to considerable challenges for end users in knowing which VEPs to use and how to use them. Here, to address these issues, we provide guidelines and recommendations for the release of novel VEPs. Emphasising open-source availability, transparent methodologies, clear variant effect score interpretations, standardised scales, accessible predictions, and rigorous training data disclosure, we aim to improve the usability and interpretability of VEPs, and promote their integration into analysis and evaluation pipelines. We also provide a large, categorised list of currently available VEPs, aiming to facilitate the discovery and encourage the usage of novel methods within the scientific community.

Empowering Protein Engineering through Recombination of Beneficial Substitutions

Directed evolution stands as a seminal technology for generating novel protein functionalities, a cornerstone in biocatalysis, metabolic engineering, and synthetic biology. Today, with the development of various mutagenesis methods and advanced analytical machines, the challenge of diversity generation and high-throughput screening platforms is largely solved, and one of the remaining challenges is: how to empower the potential of single beneficial substitutions with recombination to achieve the epistatic effect. This review overviews experimental and computer-assisted recombination methods in protein engineering campaigns. In addition, integrated and machine learning-guided strategies were highlighted to discuss how these recombination approaches contribute to generating the screening library with better diversity, coverage, and size. A decision tree was finally summarized to guide the further selection of proper recombination strategies in practice, which was beneficial for accelerating protein engineering.

Fitness Effects of Phenotypic Mutations at Proteome-Scale Reveal Optimality of Translation Machinery

Errors in protein translation can lead to non-genetic, phenotypic mutations, including amino acid misincorporations. While phenotypic mutations can increase protein diversity, the systematic characterization of their proteome-wide frequencies and their evolutionary impact has been lacking. Here, we developed a mechanistic model of translation errors to investigate how selection acts on protein populations produced by amino acid misincorporations. We fitted the model to empirical observations of misincorporations obtained from over a hundred mass spectrometry datasets of E. coli and S. cerevisiae. We found that on average 20% to 23% of proteins synthesized in the cell are expected to harbor at least one amino acid misincorporation, and that deleterious misincorporations are less likely to occur. Combining misincorporation probabilities and the estimated fitness effects of amino acid substitutions in a population genetics framework, we found 74% of mistranslation events in E. coli and 94% in S. cerevisiae to be neutral. We further show that the set of available synonymous tRNAs is subject to evolutionary pressure, as the presence of missing tRNAs would increase codon-anticodon cross-reactivity and misincorporation error rates. Overall, we find that the translation machinery is likely optimal in E. coli and S. cerevisiae and that both local solutions at the level of codons and a global solution such as the tRNA pool can mitigate the impact of translation errors. We provide a framework to study the evolutionary impact of codon-specific translation errors and a method for their proteome-wide detection across organisms and conditions.

PanEffect: a pan-genome visualization tool for variant effects in maize

SUMMARY

Understanding the effects of genetic variants is crucial for accurately predicting traits and functional outcomes. Recent approaches have utilized artificial intelligence and protein language models to score all possible missense variant effects at the proteome level for a single genome, but a reliable tool is needed to explore these effects at the pan-genome level. To address this gap, we introduce a new tool called PanEffect. We implemented PanEffect at MaizeGDB to enable a comprehensive examination of the potential effects of coding variants across 50 maize genomes. The tool allows users to visualize over 550 million possible amino acid substitutions in the B73 maize reference genome and to observe the effects of the 2.3 million natural variations in the maize pan-genome. Each variant effect score, calculated from the Evolutionary Scale Modeling (ESM) protein language model, shows the log-likelihood ratio difference between B73 and all variants in the pan-genome. These scores are shown using heatmaps spanning benign outcomes to potential functional consequences. In addition, PanEffect displays secondary structures and functional domains along with the variant effects, offering additional functional and structural context. Using PanEffect, researchers now have a platform to explore protein variants and identify genetic targets for crop enhancement.

AVAILABILITY AND IMPLEMENTATION

The PanEffect code is freely available on GitHub (https://github.com/Maize-Genetics-and-Genomics-Database/PanEffect). A maize implementation of PanEffect and underlying datasets are available at MaizeGDB (https://www.maizegdb.org/effect/maize/).

Characterizing and predicting ccRCC-causing missense mutations in Von Hippel-Lindau disease

BACKGROUND

Mutations within the Von Hippel-Lindau (VHL) tumor suppressor gene are known to cause VHL disease, which is characterized by the formation of cysts and tumors in multiple organs of the body, particularly clear cell renal cell carcinoma (ccRCC). A major challenge in clinical practice is determining tumor risk from a given mutation in the VHL gene. Previous efforts have been hindered by limited available clinical data and technological constraints.

METHODS

To overcome this, we initially manually curated the largest set of clinically validated VHL mutations to date, enabling a robust assessment of existing predictive tools on an independent test set. Additionally, we comprehensively characterized the effects of mutations within VHL using in silico biophysical tools describing changes in protein stability, dynamics and affinity to binding partners to provide insights into the structure-phenotype relationship. These descriptive properties were used as molecular features for the construction of a machine learning model, designed to predict the risk of ccRCC development as a result of a VHL missense mutation.

RESULTS

Analysis of our model showed an accuracy of 0.81 in the identification of ccRCC-causing missense mutations, and a Matthew's Correlation Coefficient of 0.44 on a non-redundant blind test, a significant improvement in comparison to the previous available approaches.

CONCLUSION

This work highlights the power of using protein 3D structure to fully explore the range of molecular and functional consequences of genomic variants. We believe this optimized model will better enable its clinical implementation and assist guiding patient risk stratification and management.

Advances in ligand-specific biosensing for structurally similar molecules

The specificity of biological systems makes it possible to develop biosensors targeting specific metabolites, toxins, and pollutants in complex medical or environmental samples without interference from structurally similar compounds. For the last two decades, great efforts have been devoted to creating proteins or nucleic acids with novel properties through synthetic biology strategies. Beyond augmenting biocatalytic activity, expanding target substrate scopes, and enhancing enzymes' enantioselectivity and stability, an increasing research area is the enhancement of molecular specificity for genetically encoded biosensors. Here, we summarize recent advances in the development of highly specific biosensor systems and their essential applications. First, we describe the rational design principles required to create libraries containing potential mutants with less promiscuity or better specificity. Next, we review the emerging high-throughput screening techniques to engineer biosensing specificity for the desired target. Finally, we examine the computer-aided evaluation and prediction methods to facilitate the construction of ligand-specific biosensors.

Machine learning and phylogenetic analysis allow for predicting antibiotic resistance in M. tuberculosis

BACKGROUND

Antimicrobial resistance (AMR) poses a significant global health threat, and an accurate prediction of bacterial resistance patterns is critical for effective treatment and control strategies. In recent years, machine learning (ML) approaches have emerged as powerful tools for analyzing large-scale bacterial AMR data. However, ML methods often ignore evolutionary relationships among bacterial strains, which can greatly impact performance of the ML methods, especially if resistance-associated features are attempted to be detected. Genome-wide association studies (GWAS) methods like linear mixed models accounts for the evolutionary relationships in bacteria, but they uncover only highly significant variants which have already been reported in literature.

RESULTS

In this work, we introduce a novel phylogeny-related parallelism score (PRPS), which measures whether a certain feature is correlated with the population structure of a set of samples. We demonstrate that PRPS can be used, in combination with SVM- and random forest-based models, to reduce the number of features in the analysis, while simultaneously increasing models' performance. We applied our pipeline to publicly available AMR data from PATRIC database for Mycobacterium tuberculosis against six common antibiotics.

CONCLUSIONS

Using our pipeline, we re-discovered known resistance-associated mutations as well as new candidate mutations which can be related to resistance and not previously reported in the literature. We demonstrated that taking into account phylogenetic relationships not only improves the model performance, but also yields more biologically relevant predicted most contributing resistance markers.

Characterization on the oncogenic effect of the missense mutations of p53 via machine learning

Dysfunctions caused by missense mutations in the tumour suppressor p53 have been extensively shown to be a leading driver of many cancers. Unfortunately, it is time-consuming and labour-intensive to experimentally elucidate the effects of all possible missense variants. Recent works presented a comprehensive dataset and machine learning model to predict the functional outcome of mutations in p53. Despite the well-established dataset and precise predictions, this tool was trained on a complicated model with limited predictions on p53 mutations. In this work, we first used computational biophysical tools to investigate the functional consequences of missense mutations in p53, informing a bias of deleterious mutations with destabilizing effects. Combining these insights with experimental assays, we present two interpretable machine learning models leveraging both experimental assays and in silico biophysical measurements to accurately predict the functional consequences on p53 and validate their robustness on clinical data. Our final model based on nine features obtained comparable predictive performance with the state-of-the-art p53 specific method and outperformed other generalized, widely used predictors. Interpreting our models revealed that information on residue p53 activity, polar atom distances and changes in p53 stability were instrumental in the decisions, consistent with a bias of the properties of deleterious mutations. Our predictions have been computed for all possible missense mutations in p53, offering clinical diagnostic utility, which is crucial for patient monitoring and the development of personalized cancer treatment.

Machine Learning-Guided Protein Engineering

Recent progress in engineering highly promising biocatalysts has increasingly involved machine learning methods. These methods leverage existing experimental and simulation data to aid in the discovery and annotation of promising enzymes, as well as in suggesting beneficial mutations for improving known targets. The field of machine learning for protein engineering is gathering steam, driven by recent success stories and notable progress in other areas. It already encompasses ambitious tasks such as understanding and predicting protein structure and function, catalytic efficiency, enantioselectivity, protein dynamics, stability, solubility, aggregation, and more. Nonetheless, the field is still evolving, with many challenges to overcome and questions to address. In this Perspective, we provide an overview of ongoing trends in this domain, highlight recent case studies, and examine the current limitations of machine learning-based methods. We emphasize the crucial importance of thorough experimental validation of emerging models before their use for rational protein design. We present our opinions on the fundamental problems and outline the potential directions for future research.

Uncovering the Molecular Drivers of NHEJ DNA Repair-Implicated Missense Variants and Their Functional Consequences

Variants in non-homologous end joining (NHEJ) DNA repair genes are associated with various human syndromes, including microcephaly, growth delay, Fanconi anemia, and different hereditary cancers. However, very little has been done previously to systematically record the underlying molecular consequences of NHEJ variants and their link to phenotypic outcomes. In this study, a list of over 2983 missense variants of the principal components of the NHEJ system, including DNA Ligase IV, DNA-PKcs, Ku70/80 and XRCC4, reported in the clinical literature, was initially collected. The molecular consequences of variants were evaluated using in silico biophysical tools to quantitatively assess their impact on protein folding, dynamics, stability, and interactions. Cancer-causing and population variants within these NHEJ factors were statistically analyzed to identify molecular drivers. A comprehensive catalog of NHEJ variants from genes known to be mutated in cancer was curated, providing a resource for better understanding their role and molecular mechanisms in diseases. The variant analysis highlighted different molecular drivers among the distinct proteins, where cancer-driving variants in anchor proteins, such as Ku70/80, were more likely to affect key protein-protein interactions, whilst those in the enzymatic components, such as DNA-PKcs, were likely to be found in intolerant regions undergoing purifying selection. We believe that the information acquired in our database will be a powerful resource to better understand the role of non-homologous end-joining DNA repair in genetic disorders, and will serve as a source to inspire other investigations to understand the disease further, vital for the development of improved therapeutic strategies.

Topological deep learning based deep mutational scanning

High-throughput deep mutational scanning (DMS) experiments have significantly impacted protein engineering, drug discovery, immunology, cancer biology, and evolutionary biology by enabling the systematic understanding of protein functions. However, the mutational space associated with proteins is astronomically large, making it overwhelming for current experimental capabilities. Therefore, alternative methods for DMS are imperative. We propose a topological deep learning (TDL) paradigm to facilitate in silico DMS. We utilize a new topological data analysis (TDA) technique based on the persistent spectral theory, also known as persistent Laplacian, to capture both topological invariants and the homotopic shape evolution of data. To validate our TDL-DMS model, we use SARS-CoV-2 datasets and show excellent accuracy and reliability for binding interface mutations. This finding is significant for SARS-CoV-2 variant forecasting and designing effective antibodies and vaccines. Our proposed model is expected to have a significant impact on drug discovery, vaccine design, precision medicine, and protein engineering.

Learning protein fitness landscapes with deep mutational scanning data from multiple sources

One of the key points of machine learning-assisted directed evolution (MLDE) is the accurate learning of the fitness landscape, a conceptual mapping from sequence variants to the desired function. Here, we describe a multi-protein training scheme that leverages the existing deep mutational scanning data from diverse proteins to aid in understanding the fitness landscape of a new protein. Proof-of-concept trials are designed to validate this training scheme in three aspects: random and positional extrapolation for single-variant effects, zero-shot fitness predictions for new proteins, and extrapolation for higher-order variant effects from single-variant effects. Moreover, our study identified previously overlooked strong baselines, and their unexpectedly good performance brings our attention to the pitfalls of MLDE. Overall, these results may improve our understanding of the association between different protein fitness profiles and shed light on developing better machine learning-assisted approaches to the directed evolution of proteins. A record of this paper's transparent peer review process is included in the supplemental information.

Updated benchmarking of variant effect predictors using deep mutational scanning

The assessment of variant effect predictor (VEP) performance is fraught with biases introduced by benchmarking against clinical observations. In this study, building on our previous work, we use independently generated measurements of protein function from deep mutational scanning (DMS) experiments for 26 human proteins to benchmark 55 different VEPs, while introducing minimal data circularity. Many top-performing VEPs are unsupervised methods including EVE, DeepSequence and ESM-1v, a protein language model that ranked first overall. However, the strong performance of recent supervised VEPs, in particular VARITY, shows that developers are taking data circularity and bias issues seriously. We also assess the performance of DMS and unsupervised VEPs for discriminating between known pathogenic and putatively benign missense variants. Our findings are mixed, demonstrating that some DMS datasets perform exceptionally at variant classification, while others are poor. Notably, we observe a striking correlation between VEP agreement with DMS data and performance in identifying clinically relevant variants, strongly supporting the validity of our rankings and the utility of DMS for independent benchmarking.

Cross-protein transfer learning substantially improves disease variant prediction

BACKGROUND

Genetic variation in the human genome is a major determinant of individual disease risk, but the vast majority of missense variants have unknown etiological effects. Here, we present a robust learning framework for leveraging saturation mutagenesis experiments to construct accurate computational predictors of proteome-wide missense variant pathogenicity.

RESULTS

We train cross-protein transfer (CPT) models using deep mutational scanning (DMS) data from only five proteins and achieve state-of-the-art performance on clinical variant interpretation for unseen proteins across the human proteome. We also improve predictive accuracy on DMS data from held-out proteins. High sensitivity is crucial for clinical applications and our model CPT-1 particularly excels in this regime. For instance, at 95% sensitivity of detecting human disease variants annotated in ClinVar, CPT-1 improves specificity to 68%, from 27% for ESM-1v and 55% for EVE. Furthermore, for genes not used to train REVEL, a supervised method widely used by clinicians, we show that CPT-1 compares favorably with REVEL. Our framework combines predictive features derived from general protein sequence models, vertebrate sequence alignments, and AlphaFold structures, and it is adaptable to the future inclusion of other sources of information. We find that vertebrate alignments, albeit rather shallow with only 100 genomes, provide a strong signal for variant pathogenicity prediction that is complementary to recent deep learning-based models trained on massive amounts of protein sequence data. We release predictions for all possible missense variants in 90% of human genes.

CONCLUSIONS

Our results demonstrate the utility of mutational scanning data for learning properties of variants that transfer to unseen proteins.

An Atlas of Variant Effects to understand the genome at nucleotide resolution

Sequencing has revealed hundreds of millions of human genetic variants, and continued efforts will only add to this variant avalanche. Insufficient information exists to interpret the effects of most variants, limiting opportunities for precision medicine and comprehension of genome function. A solution lies in experimental assessment of the functional effect of variants, which can reveal their biological and clinical impact. However, variant effect assays have generally been undertaken reactively for individual variants only after and, in most cases long after, their first observation. Now, multiplexed assays of variant effect can characterise massive numbers of variants simultaneously, yielding variant effect maps that reveal the function of every possible single nucleotide change in a gene or regulatory element. Generating maps for every protein encoding gene and regulatory element in the human genome would create an 'Atlas' of variant effect maps and transform our understanding of genetics and usher in a new era of nucleotide-resolution functional knowledge of the genome. An Atlas would reveal the fundamental biology of the human genome, inform human evolution, empower the development and use of therapeutics and maximize the utility of genomics for diagnosing and treating disease. The Atlas of Variant Effects Alliance is an international collaborative group comprising hundreds of researchers, technologists and clinicians dedicated to realising an Atlas of Variant Effects to help deliver on the promise of genomics.

Identifying the Molecular Drivers of Pathogenic Aldehyde Dehydrogenase Missense Mutations in Cancer and Non-Cancer Diseases

Human aldehyde dehydrogenases (ALDHs) comprising 19 isoenzymes play a vital role on both endogenous and exogenous aldehyde metabolism. This NAD(P)-dependent catalytic process relies on the intact structural and functional activity of the cofactor binding, substrate interaction, and the oligomerization of ALDHs. Disruptions on the activity of ALDHs, however, could result in the accumulation of cytotoxic aldehydes, which have been linked with a wide range of diseases, including both cancers as well as neurological and developmental disorders. In our previous works, we have successfully characterised the structure-function relationships of the missense variants of other proteins. We, therefore, applied a similar analysis pipeline to identify potential molecular drivers of pathogenic ALDH missense mutations. Variants data were first carefully curated and labelled as cancer-risk, non-cancer diseases, and benign. We then leveraged various computational biophysical methods to describe the changes caused by missense mutations, informing a bias of detrimental mutations with destabilising effects. Cooperating with these insights, several machine learning approaches were further utilised to investigate the combination of features, revealing the necessity of the conservation of ALDHs. Our work aims to provide important biological perspectives on pathogenic consequences of missense mutations of ALDHs, which could be invaluable resources in the development of cancer treatment.

Functional characterization of human genomic variation linked to polygenic diseases

The burden of human disease lies predominantly in polygenic diseases. Since the early 2000s, genome-wide association studies (GWAS) have identified genetic variants and loci associated with complex traits. These have ranged from variants in coding sequences to mutations in regulatory regions, such as promoters and enhancers, as well as mutations affecting mediators of mRNA stability and other downstream regulators, such as 5' and 3'-untranslated regions (UTRs), long noncoding RNA (lncRNA), and miRNA. Recent research advances in genetics have utilized a combination of computational techniques, high-throughput in vitro and in vivo screening modalities, and precise genome editing to impute the function of diverse classes of genetic variants identified through GWAS. In this review, we highlight the vastness of genomic variants associated with polygenic disease risk and address recent advances in how genetic tools can be used to functionally characterize them.

High-throughput deep learning variant effect prediction with Sequence UNET

Understanding coding mutations is important for many applications in biology and medicine but the vast mutation space makes comprehensive experimental characterisation impossible. Current predictors are often computationally intensive and difficult to scale, including recent deep learning models. We introduce Sequence UNET, a highly scalable deep learning architecture that classifies and predicts variant frequency from sequence alone using multi-scale representations from a fully convolutional compression/expansion architecture. It achieves comparable pathogenicity prediction to recent methods. We demonstrate scalability by analysing 8.3B variants in 904,134 proteins detected through large-scale proteomics. Sequence UNET runs on modest hardware with a simple Python package.

Molecular Property Prediction by Combining LSTM and GAT

Molecular property prediction is an important direction in computer-aided drug design. In this paper, to fully explore the information from SMILE stings and graph data of molecules, we combined the SALSTM and GAT methods in order to mine the feature information of molecules from sequences and graphs. The embedding atoms are obtained through SALSTM, firstly using SMILES strings, and they are combined with graph node features and fed into the GAT to extract the global molecular representation. At the same time, data augmentation is added to enlarge the training dataset and improve the performance of the model. Finally, to enhance the interpretability of the model, the attention layers of both models are fused together to highlight the key atoms. Comparison with other graph-based and sequence-based methods, for multiple datasets, shows that our method can achieve high prediction accuracy with good generalizability.

Machine Learning Guided Design of High-Affinity ACE2 Decoys for SARS-CoV-2 Neutralization

A potential therapeutic strategy for neutralizing SARS-CoV-2 infection is engineering high-affinity soluble ACE2 decoy proteins to compete for binding to the viral spike (S) protein. Previously, a deep mutational scan of ACE2 was performed and has led to the identification of a triple mutant variant, named sACE22.v.2.4, that exhibits subnanomolar affinity to the receptor-binding domain (RBD) of S. Using a recently developed transfer learning algorithm, TLmutation, we sought to identify other ACE2 variants that may exhibit similar binding affinity with decreased mutational load. Upon training a TLmutation model on the effects of single mutations, we identified multiple ACE2 double mutants that bind SARS-CoV-2 S with tighter affinity as compared to the wild type, most notably L79V;N90D that binds RBD similarly to ACE22.v.2.4. The experimental validation of the double mutants successfully demonstrates the use of machine learning approaches for engineering protein-protein interactions and identifying high-affinity ACE2 peptides for targeting SARS-CoV-2.

Using machine learning to predict the effects and consequences of mutations in proteins

Machine and deep learning approaches can leverage the increasingly available massive datasets of protein sequences, structures, and mutational effects to predict variants with improved fitness. Many different approaches are being developed, but systematic benchmarking studies indicate that even though the specifics of the machine learning algorithms matter, the more important constraint comes from the data availability and quality utilized during training. In cases where little experimental data are available, unsupervised and self-supervised pre-training with generic protein datasets can still perform well after subsequent refinement via hybrid or transfer learning approaches. Overall, recent progress in this field has been staggering, and machine learning approaches will likely play a major role in future breakthroughs in protein biochemistry and engineering.

Why does the X chromosome lag behind autosomes in GWAS findings?

The X-chromosome is among the largest human chromosomes. It differs from autosomes by a number of important features including hemizygosity in males, an almost complete inactivation of one copy in females, and unique patterns of recombination. We used data from the Catalog of Published Genome Wide Association Studies to compare densities of the GWAS-detected SNPs on the X-chromosome and autosomes. The density of GWAS-detected SNPs on the X-chromosome is 6-fold lower compared to the density of the GWAS-detected SNPs on autosomes. Differences between the X-chromosome and autosomes cannot be explained by differences in the overall SNP density, lower X-chromosome coverage by genotyping platforms or low call rate of X-chromosomal SNPs. Similar differences in the density of GWAS-detected SNPs were found in female-only GWASs (e.g. ovarian cancer GWASs). We hypothesized that the lower density of GWAS-detected SNPs on the X-chromosome compared to autosomes is not a result of a methodological bias, e.g. differences in coverage or call rates, but has a real underlying biological reason-a lower density of functional SNPs on the X-chromosome versus autosomes. This hypothesis is supported by the observation that (i) the overall SNP density of X-chromosome is lower compared to the SNP density on autosomes and that (ii) the density of genic SNPs on the X-chromosome is lower compared to autosomes while densities of intergenic SNPs are similar.

Deep mutational scanning: A versatile tool in systematically mapping genotypes to phenotypes

Unveiling how genetic variations lead to phenotypic variations is one of the key questions in evolutionary biology, genetics, and biomedical research. Deep mutational scanning (DMS) technology has allowed the mapping of tens of thousands of genetic variations to phenotypic variations efficiently and economically. Since its first systematic introduction about a decade ago, we have witnessed the use of deep mutational scanning in many research areas leading to scientific breakthroughs. Also, the methods in each step of deep mutational scanning have become much more versatile thanks to the oligo-synthesizing technology, high-throughput phenotyping methods and deep sequencing technology. However, each specific possible step of deep mutational scanning has its pros and cons, and some limitations still await further technological development. Here, we discuss recent scientific accomplishments achieved through the deep mutational scanning and describe widely used methods in each step of deep mutational scanning. We also compare these different methods and analyze their advantages and disadvantages, providing insight into how to design a deep mutational scanning study that best suits the aims of the readers' projects.

Shared Cancer Dataset Analysis Identifies and Predicts the Quantitative Effects of Pan-Cancer Somatic Driver Variants

Driver mutations endow tumors with selective advantages and produce an array of pathogenic effects. Determining the function of somatic variants is important for understanding cancer biology and identifying optimal therapies. Here, we compiled a shared dataset from several cancer genomic databases. Two measures were applied to 535 cancer genes based on observed and expected frequencies of driver variants as derived from cancer-specific rates of somatic mutagenesis. The first measure comprised a binary classifier based on a binomial test; the second was tumor variant amplitude (TVA), a continuous measure representing the selective advantage of individual variants. TVA outperformed all other computational tools in terms of its correlation with experimentally derived functional scores of cancer mutations. TVA also highly correlated with drug response, overall survival, and other clinical implications in relevant cancer genes. This study demonstrates how a selective advantage measure based on a large cancer dataset significantly impacts our understanding of the spectral effect of driver variants in cancer. The impact of this information will increase as cancer treatment becomes more precise and personalized to tumor-specific mutations.

SIGNIFICANCE

A new selective advantage estimation assists in oncogenic driver identification and relative effect measurements, enabling better prognostication, therapy selection, and prioritization.

Predicting mutational function using machine learning

Genetic variations are one of the major causes of phenotypic variations between human individuals. Although beneficial as being the substrate of evolution, germline mutations may cause diseases, including Mendelian diseases and complex diseases such as diabetes and heart diseases. Mutations occurring in somatic cells are a main cause of cancer and likely cause age-related phenotypes and other age-related diseases. Because of the high abundance of genetic variations in the human genome, i.e., millions of germline variations per human subject and thousands of additional somatic mutations per cell, it is technically challenging to experimentally verify the function of every possible mutation and their interactions. Significant progress has been made to solve this problem using computational approaches, especially machine learning (ML). Here, we review the progress and achievements made in recent years in this field of research. We classify the computational models in two ways: one according to their prediction goals including protein structural alterations, gene expression changes, and disease risks, and the other according to their methodologies, including non-machine learning methods, classical machine learning methods, and deep neural network methods. For models in each category, we discuss their architecture, prediction accuracy, and potential limitations. This review provides new insights into the applications and future directions of computational approaches in understanding the role of mutations in aging and disease.

Toward generalizable prediction of antibody thermostability using machine learning on sequence and structure features

Over the last three decades, the appeal for monoclonal antibodies (mAbs) as therapeutics has been steadily increasing as evident with FDA's recent landmark approval of the 100th mAb. Unlike mAbs that bind to single targets, multispecific biologics (msAbs) have garnered particular interest owing to the advantage of engaging distinct targets. One important modular component of msAbs is the single-chain variable fragment (scFv). Despite the exquisite specificity and affinity of these scFv modules, their relatively poor thermostability often hampers their development as a potential therapeutic drug. In recent years, engineering antibody sequences to enhance their stability by mutations has gained considerable momentum. As experimental methods for antibody engineering are time-intensive, laborious and expensive, computational methods serve as a fast and inexpensive alternative to conventional routes. In this work, we show two machine learning approaches - one with pre-trained language models (PTLM) capturing functional effects of sequence variation, and second, a supervised convolutional neural network (CNN) trained with Rosetta energetic features - to better classify thermostable scFv variants from sequence. Both of these models are trained over temperature-specific data (TS50 measurements) derived from multiple libraries of scFv sequences. On out-of-distribution (refers to the fact that the out-of-distribution sequnes are blind to the algorithm) sequences, we show that a sufficiently simple CNN model performs better than general pre-trained language models trained on diverse protein sequences (average Spearman correlation coefficient, ρ , of 0.4 as opposed to 0.15). On the other hand, an antibody-specific language model performs comparatively better than the CNN model on the same task (ρ = 0.52). Further, we demonstrate that for an independent mAb with available thermal melting temperatures for 20 experimentally characterized thermostable mutations, these models trained on TS50 data could identify 18 residue positions and 5 identical amino-acid mutations showing remarkable generalizability. Our results suggest that such models can be broadly applicable for improving the biological characteristics of antibodies. Further, transferring such models for alternative physicochemical properties of scFvs can have potential applications in optimizing large-scale production and delivery of mAbs or bsAbs.

Integrating deep mutational scanning and low-throughput mutagenesis data to predict the impact of amino acid variants

BACKGROUND

Evaluating the impact of amino acid variants has been a critical challenge for studying protein function and interpreting genomic data. High-throughput experimental methods like deep mutational scanning (DMS) can measure the effect of large numbers of variants in a target protein, but because DMS studies have not been performed on all proteins, researchers also model DMS data computationally to estimate variant impacts by predictors.

RESULTS

In this study, we extended a linear regression-based predictor to explore whether incorporating data from alanine scanning (AS), a widely used low-throughput mutagenesis method, would improve prediction results. To evaluate our model, we collected 146 AS datasets, mapping to 54 DMS datasets across 22 distinct proteins.

CONCLUSIONS

We show that improved model performance depends on the compatibility of the DMS and AS assays, and the scale of improvement is closely related to the correlation between DMS and AS results.

Reducing uncertainty in genetic testing with Saturation Genome Editing

Accurate interpretation of human genetic data is critical for optimizing outcomes in the era of genomic medicine. Powerful methods for testing genetic variants for functional effects are allowing researchers to characterize thousands of variants across disease genes. Here, we review experimental tools enabling highly scalable assays of variants, focusing specifically on Saturation Genome Editing (SGE). We discuss examples of how this technique is being implemented for variant testing at scale and describe how SGE data for BRCA1 have been clinically validated and used to aid variant interpretation. The initial success at predicting variant pathogenicity with SGE has spurred efforts to expand this and related techniques to many more genes.

Scalable Functional Assays for the Interpretation of Human Genetic Variation

Scalable sequence-function studies have enabled the systematic analysis and cataloging of hundreds of thousands of coding and noncoding genetic variants in the human genome. This has improved clinical variant interpretation and provided insights into the molecular, biophysical, and cellular effects of genetic variants at an astonishing scale and resolution across the spectrum of allele frequencies. In this review, we explore current applications and prospects for the field and outline the principles underlying scalable functional assay design, with a focus on the study of single-nucleotide coding and noncoding variants.

Intragenic compensation through the lens of deep mutational scanning

A significant fraction of mutations in proteins are deleterious and result in adverse consequences for protein function, stability, or interaction with other molecules. Intragenic compensation is a specific case of positive epistasis when a neutral missense mutation cancels effect of a deleterious mutation in the same protein. Permissive compensatory mutations facilitate protein evolution, since without them all sequences would be extremely conserved. Understanding compensatory mechanisms is an important scientific challenge at the intersection of protein biophysics and evolution. In human genetics, intragenic compensatory interactions are important since they may result in variable penetrance of pathogenic mutations or fixation of pathogenic human alleles in orthologous proteins from related species. The latter phenomenon complicates computational and clinical inference of an allele's pathogenicity. Deep mutational scanning is a relatively new technique that enables experimental studies of functional effects of thousands of mutations in proteins. We review the important aspects of the field and discuss existing limitations of current datasets. We reviewed ten published DMS datasets with quantified functional effects of single and double mutations and described rates and patterns of intragenic compensation in eight of them.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12551-022-01005-w.

High-throughput approaches to functional characterization of genetic variation in yeast

Expansion of sequencing efforts to include thousands of genomes is providing a fundamental resource for determining the genetic diversity that exists in a population. Now, high-throughput approaches are necessary to begin to understand the role these genotypic changes play in affecting phenotypic variation. Saccharomyces cerevisiae maintains its position as an excellent model system to determine the function of unknown variants with its exceptional genetic diversity, phenotypic diversity, and reliable genetic manipulation tools. Here, we review strategies and techniques developed in yeast that scale classic approaches of assessing variant function. These approaches improve our ability to better map quantitative trait loci at a higher resolution, even for rare variants, and are already providing greater insight into the role that different types of mutations play in phenotypic variation and evolution not just in yeast but across taxa.

Genome interpretation using in silico predictors of variant impact

Estimating the effects of variants found in disease driver genes opens the door to personalized therapeutic opportunities. Clinical associations and laboratory experiments can only characterize a tiny fraction of all the available variants, leaving the majority as variants of unknown significance (VUS). In silico methods bridge this gap by providing instant estimates on a large scale, most often based on the numerous genetic differences between species. Despite concerns that these methods may lack reliability in individual subjects, their numerous practical applications over cohorts suggest they are already helpful and have a role to play in genome interpretation when used at the proper scale and context. In this review, we aim to gain insights into the training and validation of these variant effect predicting methods and illustrate representative types of experimental and clinical applications. Objective performance assessments using various datasets that are not yet published indicate the strengths and limitations of each method. These show that cautious use of in silico variant impact predictors is essential for addressing genome interpretation challenges.

Embeddings from protein language models predict conservation and variant effects

The emergence of SARS-CoV-2 variants stressed the demand for tools allowing to interpret the effect of single amino acid variants (SAVs) on protein function. While Deep Mutational Scanning (DMS) sets continue to expand our understanding of the mutational landscape of single proteins, the results continue to challenge analyses. Protein Language Models (pLMs) use the latest deep learning (DL) algorithms to leverage growing databases of protein sequences. These methods learn to predict missing or masked amino acids from the context of entire sequence regions. Here, we used pLM representations (embeddings) to predict sequence conservation and SAV effects without multiple sequence alignments (MSAs). Embeddings alone predicted residue conservation almost as accurately from single sequences as ConSeq using MSAs (two-state Matthews Correlation Coefficient-MCC-for ProtT5 embeddings of 0.596 ± 0.006 vs. 0.608 ± 0.006 for ConSeq). Inputting the conservation prediction along with BLOSUM62 substitution scores and pLM mask reconstruction probabilities into a simplistic logistic regression (LR) ensemble for Variant Effect Score Prediction without Alignments (VESPA) predicted SAV effect magnitude without any optimization on DMS data. Comparing predictions for a standard set of 39 DMS experiments to other methods (incl. ESM-1v, DeepSequence, and GEMME) revealed our approach as competitive with the state-of-the-art (SOTA) methods using MSA input. No method outperformed all others, neither consistently nor statistically significantly, independently of the performance measure applied (Spearman and Pearson correlation). Finally, we investigated binary effect predictions on DMS experiments for four human proteins. Overall, embedding-based methods have become competitive with methods relying on MSAs for SAV effect prediction at a fraction of the costs in computing/energy. Our method predicted SAV effects for the entire human proteome (~ 20 k proteins) within 40 min on one Nvidia Quadro RTX 8000. All methods and data sets are freely available for local and online execution through bioembeddings.com, https://github.com/Rostlab/VESPA , and PredictProtein.

ProteinGLUE multi-task benchmark suite for self-supervised protein modeling

Self-supervised language modeling is a rapidly developing approach for the analysis of protein sequence data. However, work in this area is heterogeneous and diverse, making comparison of models and methods difficult. Moreover, models are often evaluated only on one or two downstream tasks, making it unclear whether the models capture generally useful properties. We introduce the ProteinGLUE benchmark for the evaluation of protein representations: a set of seven per-amino-acid tasks for evaluating learned protein representations. We also offer reference code, and we provide two baseline models with hyperparameters specifically trained for these benchmarks. Pre-training was done on two tasks, masked symbol prediction and next sentence prediction. We show that pre-training yields higher performance on a variety of downstream tasks such as secondary structure and protein interaction interface prediction, compared to no pre-training. However, the larger base model does not outperform the smaller medium model. We expect the ProteinGLUE benchmark dataset introduced here, together with the two baseline pre-trained models and their performance evaluations, to be of great value to the field of protein sequence-based property prediction. Availability: code and datasets from https://github.com/ibivu/protein-glue .

An intermediate-effect size variant in UMOD confers risk for chronic kidney disease

The kidney-specific gene UMOD encodes for uromodulin, the most abundant protein excreted in normal urine. Rare large-effect variants in UMOD cause autosomal dominant tubulointerstitial kidney disease (ADTKD), while common low-impact variants strongly associate with kidney function and the risk of chronic kidney disease (CKD) in the general population. It is unknown whether intermediate-effect variants in UMOD contribute to CKD. Here, candidate intermediate-effect UMOD variants were identified using large-population and ADTKD cohorts. Biological and phenotypical effects were investigated using cell models, in silico simulations, patient samples, and international databases and biobanks. Eight UMOD missense variants reported in ADTKD are present in the Genome Aggregation Database (gnomAD), with minor allele frequency (MAF) ranging from 10-5 to 10-3. Among them, the missense variant p.Thr62Pro is detected in ∼1/1,000 individuals of European ancestry, shows incomplete penetrance but a high genetic load in familial clusters of CKD, and is associated with kidney failure in the 100,000 Genomes Project (odds ratio [OR] = 3.99 [1.84 to 8.98]) and the UK Biobank (OR = 4.12 [1.32 to 12.85). Compared with canonical ADTKD mutations, the p.Thr62Pro carriers displayed reduced disease severity, with slower progression of CKD and an intermediate reduction of urinary uromodulin levels, in line with an intermediate trafficking defect in vitro and modest induction of endoplasmic reticulum (ER) stress. Identification of an intermediate-effect UMOD variant completes the spectrum of UMOD-associated kidney diseases and provides insights into the mechanisms of ADTKD and the genetic architecture of CKD.

Modeling mutational effects on biochemical phenotypes using convolutional neural networks: application to SARS-CoV-2

Deep mutational scanning (DMS) experiments have been performed on SARS-CoV-2’s spike receptor-binding domain (RBD) and human angiotensin-converting enzyme 2 (ACE2) zinc-binding peptidase domain—both central players in viral infection and evolution and antibody evasion—quantifying how mutations impact biochemical phenotypes. We modeled biochemical phenotypes from massively parallel assays, using neural networks trained on protein sequence mutations in the virus and human host. Neural networks were significantly predictive of binding affinity, protein expression, and antibody escape, learning complex interactions and higher-order features that are difficult to capture with conventional methods from structural biology. Integrating the physicochemical properties of amino acids, such as hydrophobicity and long-range non-bonded energy per atom, significantly improved prediction (empirical p < 0.01). We observed concordance of the neural network predictions with molecular dynamics (multiple 500 ns or 1 μs all-atom) simulations of the spike protein-ACE2 interface, with critical implications for the use of deep learning to dissect molecular mechanisms.

•

Deep learning models of biochemical phenotypes from deep mutational scanning (DMS) data

•

Prediction performance gain from using physicochemical properties of amino acids

•

Concordance of neural network predictions with molecular dynamics simulations

•

Improved causal inference properties for neural-network-defined phenotypes

Loss-of-function, gain-of-function and dominant-negative mutations have profoundly different effects on protein structure

Most known pathogenic mutations occur in protein-coding regions of DNA and change the way proteins are made. Taking protein structure into account has therefore provided great insight into the molecular mechanisms underlying human genetic disease. While there has been much focus on how mutations can disrupt protein structure and thus cause a loss of function (LOF), alternative mechanisms, specifically dominant-negative (DN) and gain-of-function (GOF) effects, are less understood. Here, we investigate the protein-level effects of pathogenic missense mutations associated with different molecular mechanisms. We observe striking differences between recessive vs dominant, and LOF vs non-LOF mutations, with dominant, non-LOF disease mutations having much milder effects on protein structure, and DN mutations being highly enriched at protein interfaces. We also find that nearly all computational variant effect predictors, even those based solely on sequence conservation, underperform on non-LOF mutations. However, we do show that non-LOF mutations could potentially be identified by their tendency to cluster in three-dimensional space. Overall, our work suggests that many pathogenic mutations that act via DN and GOF mechanisms are likely being missed by current variant prioritisation strategies, but that there is considerable scope to improve computational predictions through consideration of molecular disease mechanisms.

Most known pathogenic mutations occur in protein-coding regions of DNA and change the way proteins are made. Here the authors analyse the locations of thousands of human disease mutations and their predicted effects on protein structure and show that,while loss-of-function mutations tend to be highly disruptive, non-loss-of-function mutations are in general much milder at a protein structural level.

How Functional Genomics Can Keep Pace With VUS Identification

Over the last two decades, an exponentially expanding number of genetic variants have been identified associated with inherited cardiac conditions. These tremendous gains also present challenges in deciphering the clinical relevance of unclassified variants or variants of uncertain significance (VUS). This review provides an overview of the advancements (and challenges) in functional and computational approaches to characterize variants and help keep pace with VUS identification related to inherited heart diseases.

GigaAssay - An adaptable high-throughput saturation mutagenesis assay platform

High-throughput assay systems have had a large impact on understanding the mechanisms of basic cell functions. However, high-throughput assays that directly assess molecular functions are limited. Herein, we describe the "GigaAssay", a modular high-throughput one-pot assay system for measuring molecular functions of thousands of genetic variants at once. In this system, each cell was infected with one virus from a library encoding thousands of Tat mutant proteins, with each viral particle encoding a random unique molecular identifier (UMI). We demonstrate proof of concept by measuring transcription of a GFP reporter in an engineered reporter cell line driven by binding of the HIV Tat transcription factor to the HIV long terminal repeat. Infected cells were flow-sorted into 3 bins based on their GFP fluorescence readout. The transcriptional activity of each Tat mutant was calculated from the ratio of signals from each bin. The use of UMIs in the GigaAssay produced a high average accuracy (95%) and positive predictive value (98%) determined by comparison to literature benchmark data, known C-terminal truncations, and blinded independent mutant tests. Including the substitution tolerance with structure/function analysis shows restricted substitution types spatially concentrated in the Cys-rich region. Tat has abundant intragenic epistasis (10%) when single and double mutants are compared.

Towards generalizable predictions for G protein-coupled receptor variant expression

Missense mutations that compromise the plasma membrane expression (PME) of integral membrane proteins are the root cause of numerous genetic diseases. Differentiation of this class of mutations from those that specifically modify the activity of the folded protein has proven useful for the development and targeting of precision therapeutics. Nevertheless, it remains challenging to predict the effects of mutations on the stability and/ or expression of membrane proteins. In this work, we utilize deep mutational scanning data to train a series of artificial neural networks to predict the PME of transmembrane domain variants of G protein-coupled receptors from structural and/ or evolutionary features. We show that our best-performing network, which we term the PME predictor, can recapitulate mutagenic trends within rhodopsin and can differentiate pathogenic transmembrane domain variants that cause it to misfold from those that compromise its signaling. This network also generates statistically significant predictions for the relative PME of transmembrane domain variants for another class A G protein-coupled receptor (β₂ adrenergic receptor) but not for an unrelated voltage-gated potassium channel (KCNQ1). Notably, our analyses of these networks suggest structural features alone are generally sufficient to recapitulate the observed mutagenic trends. Moreover, our findings imply that networks trained in this manner may be generalizable to proteins that share a common fold. Implications of our findings for the design of mechanistically specific genetic predictors are discussed.

Interpreting protein variant effects with computational predictors and deep mutational scanning

Computational predictors of genetic variant effect have advanced rapidly in recent years. These programs provide clinical and research laboratories with a rapid and scalable method to assess the likely impacts of novel variants. However, it can be difficult to know to what extent we can trust their results. To benchmark their performance, predictors are often tested against large datasets of known pathogenic and benign variants. These benchmarking data may overlap with the data used to train some supervised predictors, which leads to data re-use or circularity, resulting in inflated performance estimates for those predictors. Furthermore, new predictors are usually found by their authors to be superior to all previous predictors, which suggests some degree of computational bias in their benchmarking. Large-scale functional assays known as deep mutational scans provide one possible solution to this problem, providing independent datasets of variant effect measurements. In this Review, we discuss some of the key advances in predictor methodology, current benchmarking strategies and how data derived from deep mutational scans can be used to overcome the issue of data circularity. We also discuss the ability of such functional assays to directly predict clinical impacts of mutations and how this might affect the future need for variant effect predictors.

Recent Advances in Machine Learning Variant Effect Prediction Tools for Protein Engineering

Proteins are Nature's molecular machinery and comprise diverse roles while consisting of chemically similar building blocks. In recent years, protein engineering and design have become important research areas, with many applications in the pharmaceutical, energy, and biocatalysis fields, among others-where the aim is to ultimately create a protein given desired structural and functional properties. It is often critical to model the relationship between a protein's sequence, folded structure, and biological function to assist in such protein engineering pursuits. However, significant challenges remain in concretely mapping an amino acid sequence to specific protein properties and biological activities. Mutations may enhance or diminish molecular protein function, and the epistatic interactions between mutations result in an inherently complex mapping between genetic modifications and protein function. Therefore, estimating the quantitative effects of mutations on protein function(s) remains a grand challenge of biology, bioinformatics, and many related fields and would rapidly accelerate protein engineering tasks when successful. Such estimation is often known as variant effect prediction (VEP). However, progress has been demonstrated in recent years with the development of machine learning (ML) methods in modeling the relationship between mutations and protein function. In this Review, recent advances in variant effect prediction (VEP) are discussed as tools for protein engineering, focusing on techniques incorporating gains from the broader ML community and challenges in estimating biomolecular functional differences. Primary developments highlighted include convolutional neural networks, graph neural networks, and natural language embeddings for protein sequences.

Integration of machine learning with computational structural biology of plants

Computational structural biology of proteins has developed rapidly in recent decades with the development of new computational tools and the advancement of computing hardware. However, while these techniques have widely been used to make advancements in human medicine, these methods have seen less utilization in the plant sciences. In the last several years, machine learning methods have gained popularity in computational structural biology. These methods have enabled the development of new tools which are able to address the major challenges that have hampered the wide adoption of the computational structural biology of plants. This perspective examines the remaining challenges in computational structural biology and how the development of machine learning techniques enables more in-depth computational structural biology of plants.

Spatial covariance analysis reveals the residue-by-residue thermodynamic contribution of variation to the CFTR fold

Although the impact of genome variation on the thermodynamic properties of function on the protein fold has been studied in vitro, it remains a challenge to assign these relationships across the entire polypeptide sequence in vivo. Using the Gaussian process regression based principle of Spatial CoVariance, we globally assign on a residue-by-residue basis the biological thermodynamic properties that contribute to the functional fold of CFTR in the cell. We demonstrate the existence of a thermodynamically sensitive region of the CFTR fold involving the interface between NBD1 and ICL4 that contributes to its export from endoplasmic reticulum. At the cell surface a new set of residues contribute uniquely to the management of channel function. These results support a general 'quality assurance' view of global protein fold management as an SCV principle describing the differential pre- and post-ER residue interactions contributing to compartmentalization of the energetics of the protein fold for function. Our results set the stage for future analyses of the quality systems managing protein sequence-to-function-to-structure broadly encompassing genome design leading to protein function in complex cellular relationships responsible for diversity and fitness in biology in response to the environment.

Computational and experimental methods for classifying variants of unknown clinical significance

The increase in sequencing capacity, reduction in costs, and national and international coordinated efforts have led to the widespread introduction of next-generation sequencing (NGS) technologies in patient care. More generally, human genetics and genomic medicine are gaining importance for more and more patients. Some communities are already discussing the prospect of sequencing each individual's genome at time of birth. Together with digital health records, this shall enable individualized treatments and preventive measures, so-called precision medicine. A central step in this process is the identification of disease causal mutations or variant combinations that make us more susceptible for diseases. Although various technological advances have improved the identification of genetic alterations, the interpretation and ranking of the identified variants remains a major challenge. Based on our knowledge of molecular processes or previously identified disease variants, we can identify potentially functional genetic variants and, using different lines of evidence, we are sometimes able to demonstrate their pathogenicity directly. However, the vast majority of variants are classified as variants of uncertain clinical significance (VUSs) with not enough experimental evidence to determine their pathogenicity. In these cases, computational methods may be used to improve the prioritization and an increasing toolbox of experimental methods is emerging that can be used to assay the molecular effects of VUSs. Here, we discuss how computational and experimental methods can be used to create catalogs of variant effects for a variety of molecular and cellular phenotypes. We discuss the prospects of integrating large-scale functional data with machine learning and clinical knowledge for the development of accurate pathogenicity predictions for clinical applications.