A generalized Kirkwood implicit solvent for the polarizable AMOEBA protein model.

Computational simulation of biomolecules can provide important insights into protein design, protein-ligand binding interactions, and ab initio biomolecular folding, among other applications. Accurate treatment of the solvent environment is essential in such applications, but the use of explicit solvents can add considerable cost. Implicit treatment of solvent effects using a dielectric continuum model is an attractive alternative to explicit solvation since it is able to describe solvation effects without the inclusion of solvent degrees of freedom. Previously, we described the development and parameterization of implicit solvent models for small molecules. Here, we extend the parameterization of the generalized Kirkwood (GK) implicit solvent model for use with biomolecules described by the AMOEBA force field via the addition of corrections to the calculation of effective radii that account for interstitial spaces that arise within biomolecules. These include element-specific pairwise descreening scale factors, a short-range neck contribution to describe the solvent-excluded space between pairs of nearby atoms, and finally tanh-based rescaling of the overall descreening integral. We then apply the AMOEBA/GK implicit solvent to a set of ten proteins and achieve an average coordinate root mean square deviation for the experimental structures of 2.0 Å across 500 ns simulations. Overall, the continued development of implicit solvent models will help facilitate the simulation of biomolecules on mechanistically relevant timescales.

Assessing variants of uncertain significance implicated in hearing loss using a comprehensive deafness proteome.

Hearing loss is the leading sensory deficit, affecting ~ 5% of the population. It exhibits remarkable heterogeneity across 223 genes with 6328 pathogenic missense variants, making deafness-specific expertise a prerequisite for ascribing phenotypic consequences to genetic variants. Deafness-implicated variants are curated in the Deafness Variation Database (DVD) after classification by a genetic hearing loss expert panel and thorough informatics pipeline. However, seventy percent of the 128,167 missense variants in the DVD are "variants of uncertain significance" (VUS) due to insufficient evidence for classification. Here, we use the deep learning protein prediction algorithm, AlphaFold2, to curate structures for all DVD genes. We refine these structures with global optimization and the AMOEBA force field and use DDGun3D to predict folding free energy differences (∆∆GFold) for all DVD missense variants. We find that 5772 VUSs have a large, destabilizing ∆∆GFold that is consistent with pathogenic variants. When also filtered for CADD scores (> 25.7), we determine 3456 VUSs are likely pathogenic at a probability of 99.0%. Of the 224 genes in the DVD, 166 genes (74%) exhibit one or more missense variants predicted to cause a pathogenic change in protein folding stability. The VUSs prioritized here affect 119 patients (~ 3% of cases) sequenced by the OtoSCOPE targeted panel. Approximately half of these patients previously received an inconclusive report, and reclassification of these VUSs as pathogenic provides a new genetic diagnosis for six patients.

Assessing Variants of Uncertain Significance Implicated in Hearing Loss Using a Comprehensive Deafness Proteome.

Hearing loss is the leading sensory deficit, affecting ~ 5% of the population. It exhibits remarkable heterogeneity across 223 genes with 6,328 pathogenic missense variants, making deafness-specific expertise a prerequisite for ascribing phenotypic consequences to genetic variants. Deafness-implicated variants are curated in the Deafness Variation Database (DVD) after classification by a genetic hearing loss expert panel and thorough informatics pipeline. However, seventy percent of the 128,167 missense variants in the DVD are "variants of uncertain significance" (VUS) due to insufficient evidence for classification. Here, we use the deep learning protein prediction algorithm, AlphaFold2, to curate structures for all DVD genes. We refine these structures with global optimization and the AMOEBA force field and use DDGun3D to predict folding free energy differences (∆∆G Fold ) for all DVD missense variants. We find that 5,772 VUSs have a large, destabilizing ∆∆G Fold that is consistent with pathogenic variants. When also filtered for CADD scores (> 25.7), we determine 3,456 VUSs are likely pathogenic at a probability of 99.0%. These VUSs affect 119 patients (~ 3% of cases) sequenced by the OtoSCOPE targeted panel. Approximately half of these patients previously received an inconclusive report, and reclassification of these VUSs as pathogenic provides a new genetic diagnosis for six patients.

AudioGene: refining the natural history of KCNQ4, GSDME, WFS1, and COCH-associated hearing loss.

Autosomal dominant non-syndromic hearing loss (ADNSHL) displays gene-specific progression of hearing loss, which is amenable to sequential audioprofiling. We sought to refine the natural history of ADNSHL by examining audiometric data in 5-year increments. 2175 audiograms were included from four genetic causes of ADNSHL-KCNQ4 (DFNA2), GSDME (DFNA5), WFS1 (DFNA6/14/38), and COCH (DFNA9). Annual threshold deterioration (ATD) was calculated for each gene: for the speech-frequency pure tone average, the ATD, respectively, was 0.72 dB/year, 0.94 dB/year, 0.53 dB/year, and 1.41 dB/year, with the largest drops occurring from ages 45-50 (0.89 dB/year; KCNQ4), 5-10 (1.42 dB/year; GSDME), 40-45 (0.83 dB/year; WFS1), and 50-55 (2.09 dB/year; COCH). 5-year interval analysis of audiograms reveals the gene specific natural history of KCNQ4, GSDME, WFS1 and COCH-related progressive hearing loss. Identifying ages at which hearing loss is most rapid informs clinical care and patient expectations. Natural history data are also essential to define outcomes of clinical trials that test novel therapies designed to correct or ameliorate these genetic forms of hearing loss.

Implicit Solvents for the Polarizable Atomic Multipole AMOEBA Force Field.

Computational protein design, ab initio protein/RNA folding, and protein-ligand screening can be too computationally demanding for explicit treatment of solvent. For these applications, implicit solvent offers a compelling alternative, which we describe here for the polarizable atomic multipole AMOEBA force field based on three treatments of continuum electrostatics: numerical solutions to the nonlinear and linearized versions of the Poisson-Boltzmann equation (PBE), the domain-decomposition conductor-like screening model (ddCOSMO) approximation to the PBE, and the analytic generalized Kirkwood (GK) approximation. The continuum electrostatics models are combined with a nonpolar estimator based on novel cavitation and dispersion terms. Electrostatic model parameters are numerically optimized using a least-squares style target function based on a library of 103 small-molecule solvation free energy differences. Mean signed errors for the adaptive Poisson-Boltzmann solver (APBS), ddCOSMO, and GK models are 0.05, 0.00, and 0.00 kcal/mol, respectively, while the mean unsigned errors are 0.70, 0.63, and 0.58 kcal/mol, respectively. Validation of the electrostatic response of the resulting implicit solvents, which are available in the Tinker (or Tinker-HP), OpenMM, and Force Field X software packages, is based on comparisons to explicit solvent simulations for a series of proteins and nucleic acids. Overall, the emergence of performative implicit solvent models for polarizable force fields opens the door to their use for folding and design applications.

A comparative analysis of genetic hearing loss phenotypes in European/American and Japanese populations.

We present detailed comparative analyses to assess population-level differences in patterns of genetic deafness between European/American and Japanese cohorts with non-syndromic hearing loss. One thousand eighty-three audiometric test results (921 European/American and 162 Japanese) from members of 168 families (48 European/American and 120 Japanese) with non-syndromic hearing loss secondary to pathogenic variants in one of three genes (KCNQ4, TECTA, WFS1) were studied. Audioprofile characteristics, specific mutation types, and protein domains were considered in the comparative analyses. Our findings support differences in audioprofiles driven by both mutation type (non-truncating vs. truncating) and ethnic background. The former finding confirms data that ascribe a phenotypic consequence to different mutation types in KCNQ4; the latter finding suggests that there are ethnic-specific effects (genetic and/or environmental) that impact gene-specific audioprofiles for TECTA and WFS1. Identifying the drivers of ethnic differences will refine our understanding of phenotype-genotype relationships and the biology of hearing and deafness.

Structural Insights into Hearing Loss Genetics from Polarizable Protein Repacking.

Hearing loss is associated with ∼8100 mutations in 152 genes, and within the coding regions of these genes are over 60,000 missense variants. The majority of these variants are classified as "variants of uncertain significance" to reflect our inability to ascribe a phenotypic effect to the observed amino acid change. A promising source of pathogenicity information is biophysical simulation, although input protein structures often contain defects because of limitations in experimental data and/or only distant homology to a template. Here, we combine the polarizable atomic multipole optimized energetics for biomolecular applications force field, many-body optimization theory, and graphical processing unit acceleration to repack all deafness-associated proteins and thereby improve average structure MolProbity score from 2.2 to 1.0. We then used these optimized wild-type models to create over 60,000 structures for missense variants in the Deafness Variation Database, which are being incorporated into the Deafness Variation Database to inform deafness pathogenicity prediction. Finally, this work demonstrates that advanced polarizable atomic multipole force fields are efficient enough to repack the entire human proteome.

Machine learning with the TCGA-HNSC dataset: improving usability by addressing inconsistency, sparsity, and high-dimensionality.

BACKGROUND: In the era of precision oncology and publicly available datasets, the amount of information available for each patient case has dramatically increased. From clinical variables and PET-CT radiomics measures to DNA-variant and RNA expression profiles, such a wide variety of data presents a multitude of challenges. Large clinical datasets are subject to sparsely and/or inconsistently populated fields. Corresponding sequencing profiles can suffer from the problem of high-dimensionality, where making useful inferences can be difficult without correspondingly large numbers of instances. In this paper we report a novel deployment of machine learning techniques to handle data sparsity and high dimensionality, while evaluating potential biomarkers in the form of unsupervised transformations of RNA data. We apply preprocessing, MICE imputation, and sparse principal component analysis (SPCA) to improve the usability of more than 500 patient cases from the TCGA-HNSC dataset for enhancing future oncological decision support for Head and Neck Squamous Cell Carcinoma (HNSCC). RESULTS: Imputation was shown to improve prognostic ability of sparse clinical treatment variables. SPCA transformation of RNA expression variables reduced runtime for RNA-based models, though changes to classifier performance were not significant. Gene ontology enrichment analysis of gene sets associated with individual sparse principal components (SPCs) are also reported, showing that both high- and low-importance SPCs were associated with cell death pathways, though the high-importance gene sets were found to be associated with a wider variety of cancer-related biological processes. CONCLUSIONS: MICE imputation allowed us to impute missing values for clinically informative features, improving their overall importance for predicting two-year recurrence-free survival by incorporating variance from other clinical variables. Dimensionality reduction of RNA expression profiles via SPCA reduced both computation cost and model training/evaluation time without affecting classifier performance, allowing researchers to obtain experimental results much more quickly. SPCA simultaneously provided a convenient avenue for consideration of biological context via gene ontology enrichment analysis.

Whole-gene sequencing investigation of SAT1 in attempted suicide.

Suicidal behavior imposes a tremendous cost, with current US estimates reporting approximately 1.3 million suicide attempts and more than 40,000 suicide deaths each year. Several recent research efforts have identified an association between suicidal behavior and the expression level of the spermidine/spermine N1-acetyltransferase 1 (SAT1) gene. To date, several SAT1 genetic variants have been inconsistently associated with altered gene expression and/or directly with suicidal behavior. To clarify the role SAT1 genetic variation plays in suicidal behavior risk, we present a whole-gene sequencing effort of SAT1 in 476 bipolar disorder subjects with a history of suicide attempt and 473 subjects with bipolar disorder but no suicide attempts. Agilent SureSelect target enrichment was used to sequence all exons, introns, promoter regions, and putative regulatory regions identified from the ENCODE project within 10 kb of SAT1. Individual variant, haplotype, and collapsing variant tests were performed. Our results identified no variant or assessed region of SAT1 that showed a significant association with attempted suicide, nor did any assessment show evidence for replication of previously reported associations. Overall, no evidence for SAT1 sequence variation contributing to the risk for attempted suicide could be identified. It is possible that past associations of SAT1 expression with suicidal behavior arise from variation not captured in this study, or that causal variants in the region are too rare to be detected within our sample. Larger sample sizes and broader sequencing efforts will likely be required to identify the source of SAT1 expression level associations with suicidal behavior. © 2016 Wiley Periodicals, Inc.

Audioprofile Surfaces: The 21st Century Audiogram.

OBJECTIVE: To present audiometric data in 3 dimensions by considering age as an addition dimension. METHODS: Audioprofile surfaces (APSs) were fitted to a set of audiograms by plotting each measurement of an audiogram as an independent point in 3 dimensions with the x, y, and z axes representing frequency, hearing loss in dB, and age, respectively. RESULTS: Using the Java-based APS viewer as a standalone application, APSs were pre-computed for 34 loci. By selecting APSs for the appropriate genetic locus, a clinician can compare this APS-generated average surface to a specific patient's audiogram. CONCLUSION: Audioprofile surfaces provide an easily interpreted visual representation of a person's hearing acuity relative to others with the same genetic cause of hearing loss. Audioprofile surfaces will support the generation and testing of sophisticated hypotheses to further refine our understanding of the biology of hearing.

CNV-ROC: A cost effective, computer-aided analytical performance evaluator of chromosomal microarrays.

Chromosomal microarrays (CMAs) are routinely used in both research and clinical laboratories; yet, little attention has been given to the estimation of genome-wide true and false negatives during the assessment of these assays and how such information could be used to calibrate various algorithmic metrics to improve performance. Low-throughput, locus-specific methods such as fluorescence in situ hybridization (FISH), quantitative PCR (qPCR), or multiplex ligation-dependent probe amplification (MLPA) preclude rigorous calibration of various metrics used by copy number variant (CNV) detection algorithms. To aid this task, we have established a comparative methodology, CNV-ROC, which is capable of performing a high throughput, low cost, analysis of CMAs that takes into consideration genome-wide true and false negatives. CNV-ROC uses a higher resolution microarray to confirm calls from a lower resolution microarray and provides for a true measure of genome-wide performance metrics at the resolution offered by microarray testing. CNV-ROC also provides for a very precise comparison of CNV calls between two microarray platforms without the need to establish an arbitrary degree of overlap. Comparison of CNVs across microarrays is done on a per-probe basis and receiver operator characteristic (ROC) analysis is used to calibrate algorithmic metrics, such as log2 ratio threshold, to enhance CNV calling performance. CNV-ROC addresses a critical and consistently overlooked aspect of analytical assessments of genome-wide techniques like CMAs which is the measurement and use of genome-wide true and false negative data for the calculation of performance metrics and comparison of CNV profiles between different microarray experiments.

Utilizing ethnic-specific differences in minor allele frequency to recategorize reported pathogenic deafness variants.

Ethnic-specific differences in minor allele frequency impact variant categorization for genetic screening of nonsyndromic hearing loss (NSHL) and other genetic disorders. We sought to evaluate all previously reported pathogenic NSHL variants in the context of a large number of controls from ethnically distinct populations sequenced with orthogonal massively parallel sequencing methods. We used HGMD, ClinVar, and dbSNP to generate a comprehensive list of reported pathogenic NSHL variants and re-evaluated these variants in the context of 8,595 individuals from 12 populations and 6 ethnically distinct major human evolutionary phylogenetic groups from three sources (Exome Variant Server, 1000 Genomes project, and a control set of individuals created for this study, the OtoDB). Of the 2,197 reported pathogenic deafness variants, 325 (14.8%) were present in at least one of the 8,595 controls, indicating a minor allele frequency (MAF) > 0.00006. MAFs ranged as high as 0.72, a level incompatible with pathogenicity for a fully penetrant disease like NSHL. Based on these data, we established MAF thresholds of 0.005 for autosomal-recessive variants (excluding specific variants in GJB2) and 0.0005 for autosomal-dominant variants. Using these thresholds, we recategorized 93 (4.2%) of reported pathogenic variants as benign. Our data show that evaluation of reported pathogenic deafness variants using variant MAFs from multiple distinct ethnicities and sequenced by orthogonal methods provides a powerful filter for determining pathogenicity. The proposed MAF thresholds will facilitate clinical interpretation of variants identified in genetic testing for NSHL. All data are publicly available to facilitate interpretation of genetic variants causing deafness.

Cordova: web-based management of genetic variation data.

UNLABELLED: Cordova is an out-of-the-box solution for building and maintaining an online database of genetic variations integrated with pathogenicity prediction results from popular algorithms. Our primary motivation for developing this system is to aid researchers and clinician-scientists in determining the clinical significance of genetic variations. To achieve this goal, Cordova provides an interface to review and manually or computationally curate genetic variation data as well as share it for clinical diagnostics and the advancement of research. AVAILABILITY AND IMPLEMENTATION: Cordova is open source under the MIT license and is freely available for download at https://github.com/clcg/cordova.

Advancing genetic testing for deafness with genomic technology.

BACKGROUND: Non-syndromic hearing loss (NSHL) is the most common sensory impairment in humans. Until recently its extreme genetic heterogeneity precluded comprehensive genetic testing. Using a platform that couples targeted genomic enrichment (TGE) and massively parallel sequencing (MPS) to sequence all exons of all genes implicated in NSHL, we tested 100 persons with presumed genetic NSHL and in so doing established sequencing requirements for maximum sensitivity and defined MPS quality score metrics that obviate Sanger validation of variants. METHODS: We examined DNA from 100 sequentially collected probands with presumed genetic NSHL without exclusions due to inheritance, previous genetic testing, or type of hearing loss. We performed TGE using post-capture multiplexing in variable pool sizes followed by Illumina sequencing. We developed a local Galaxy installation on a high performance computing cluster for bioinformatics analysis. RESULTS: To obtain maximum variant sensitivity with this platform 3.2-6.3 million total mapped sequencing reads per sample were required. Quality score analysis showed that Sanger validation was not required for 95% of variants. Our overall diagnostic rate was 42%, but this varied by clinical features from 0% for persons with asymmetric hearing loss to 56% for persons with bilateral autosomal recessive NSHL. CONCLUSIONS: These findings will direct the use of TGE and MPS strategies for genetic diagnosis for NSHL. Our diagnostic rate highlights the need for further research on genetic deafness focused on novel gene identification and an improved understanding of the role of non-exonic mutations. The unsolved families we have identified provide a valuable resource to address these areas.

Computational identification of operon-like transcriptional loci in eukaryotes.

Operons are primarily a bacterial phenomenon, not commonly observed in eukaryotes. However, new research indicates that operons are found in higher organisms as well. There are instances of operons found in C. elegans, Drosophila melanogaster and other eukaryotic species. We developed a prototype using positional, structural and gene expression information to identify candidate operons. We focused our efforts on "trans-spliced" operons in which the pre-mRNA is trans-spliced into individual transcripts and subsequently translated, as widely observed in C. elegans and some instances in Drosophila. We identify several candidate operons in Drosophila melanogaster of which two have been subsequently molecularly validated.

Prioritization of retinal disease genes: an integrative approach.

The discovery of novel disease-associated variations in genes is often a daunting task in highly heterogeneous disease classes. We seek a generalizable algorithm that integrates multiple publicly available genomic data sources in a machine-learning model for the prioritization of candidates identified in patients with retinal disease. To approach this problem, we generate a set of feature vectors from publicly available microarray, RNA-seq, and ChIP-seq datasets of biological relevance to retinal disease, to observe patterns in gene expression specificity among tissues of the body and the eye, in addition to photoreceptor-specific signals by the CRX transcription factor. Using these features, we describe a novel algorithm, positive and unlabeled learning for prioritization (PULP). This article compares several popular supervised learning techniques as the regression function for PULP. The results demonstrate a highly significant enrichment for previously characterized disease genes using a logistic regression method. Finally, a comparison of PULP with the popular gene prioritization tool ENDEAVOUR shows superior prioritization of retinal disease genes from previous studies. The java source code, compiled binary, assembled feature vectors, and instructions are available online at https://github.com/ahwagner/PULP.

Exon-level expression profiling of ocular tissues.

The normal gene expression profiles of the tissues in the eye are a valuable resource for considering genes likely to be involved with disease processes. We profiled gene expression in ten ocular tissues from human donor eyes using Affymetrix Human Exon 1.0 ST arrays. Ten different tissues were obtained from six different individuals and RNA was pooled. The tissues included: retina, optic nerve head (ONH), optic nerve (ON), ciliary body (CB), trabecular meshwork (TM), sclera, lens, cornea, choroid/retinal pigment epithelium (RPE) and iris. Expression values were compared with publically available Expressed Sequence Tag (EST) and RNA-sequencing resources. Known tissue-specific genes were examined and they demonstrated correspondence of expression with the representative ocular tissues. The estimated gene and exon level abundances are available online at the Ocular Tissue Database.

A genome-wide association study for primary open angle glaucoma and macular degeneration reveals novel Loci.

Glaucoma and age-related macular degeneration (AMD) are the two leading causes of visual loss in the United States. We utilized a novel study design to perform a genome-wide association for both primary open angle glaucoma (POAG) and AMD. This study design utilized a two-stage process for hypothesis generation and validation, in which each disease cohort was utilized as a control for the other. A total of 400 POAG patients and 400 AMD patients were ascertained and genotyped at 500,000 loci. This study identified a novel association of complement component 7 (C7) to POAG. Additionally, an association of central corneal thickness, a known risk factor for POAG, was found to be associated with ribophorin II (RPN2). Linked monogenic loci for POAG and AMD were also evaluated for evidence of association, none of which were found to be significantly associated. However, several yielded putative associations requiring validation. Our data suggest that POAG is more genetically complex than AMD, with no common risk alleles of large effect.

AudioGene: predicting hearing loss genotypes from phenotypes to guide genetic screening.

Autosomal dominant nonsyndromic hearing loss (ADNSHL) is a common and often progressive sensory deficit. ADNSHL displays a high degree of genetic heterogeneity and varying rates of progression. Accurate, comprehensive, and cost-effective genetic testing facilitates genetic counseling and provides valuable prognostic information to affected individuals. In this article, we describe the algorithm underlying AudioGene, a software system employing machine-learning techniques that utilizes phenotypic information derived from audiograms to predict the genetic cause of hearing loss in persons segregating ADNSHL. Our data show that AudioGene has an accuracy of 68% in predicting the causative gene within its top three predictions, as compared with 44% for a majority classifier. We also show that AudioGene remains effective for audiograms with high levels of clinical measurement noise. We identify audiometric outliers for each genetic locus and hypothesize that outliers may reflect modifying genetic effects. As personalized genomic medicine becomes more common, AudioGene will be increasingly useful as a phenotypic filter to assess pathogenicity of variants identified by massively parallel sequencing.

Prediction of cochlear implant performance by genetic mutation: the spiral ganglion hypothesis.

BACKGROUND: Up to 7% of patients with severe-to-profound deafness do not benefit from cochlear implantation. Given the high surgical implantation and clinical management cost of cochlear implantation (>$1 million lifetime cost), prospective identification of the worst performers would reduce unnecessary procedures and healthcare costs. Because cochlear implants bypass the membranous labyrinth but rely on the spiral ganglion for functionality, we hypothesize that cochlear implant (CI) performance is dictated in part by the anatomic location of the cochlear pathology that underlies the hearing loss. As a corollary, we hypothesize that because genetic testing can identify sites of cochlear pathology, it may be useful in predicting CI performance. METHODS: 29 adult CI recipients with idiopathic adult-onset severe-to-profound hearing loss were studied. DNA samples were subjected to solution-based sequence capture and massively parallel sequencing using the OtoSCOPE(®) platform. The cohort was divided into three CI performance groups (good, intermediate, poor) and genetic causes of deafness were correlated with audiometric data to determine whether there was a gene-specific impact on CI performance. RESULTS: The genetic cause of deafness was determined in 3/29 (10%) individuals. The two poor performers segregated mutations in TMPRSS3, a gene expressed in the spiral ganglion, while the good performer segregated mutations in LOXHD1, a gene expressed in the membranous labyrinth. Comprehensive literature review identified other good performers with mutations in membranous labyrinth-expressed genes; poor performance was associated with spiral ganglion-expressed genes. CONCLUSIONS: Our data support the underlying hypothesis that mutations in genes preferentially expressed in the spiral ganglion portend poor CI performance while mutations in genes expressed in the membranous labyrinth portend good CI performance. Although the low mutation rate in known deafness genes in this cohort likely relates to the ascertainment characteristics (postlingual hearing loss in adult CI recipients), these data suggest that genetic testing should be implemented as part of the CI evaluation to test this association prospectively.