How to evaluate performance of prediction methods? Measures and their interpretation in variation effect analysis

PON-tstab: Protein Variant Stability Predictor. Importance of Training Data Quality

Several methods have been developed to predict effects of amino acid substitutions on protein stability. Benchmark datasets are essential for method training and testing and have numerous requirements including that the data is representative for the investigated phenomenon. Available machine learning algorithms for variant stability have all been trained with ProTherm data. We noticed a number of issues with the contents, quality and relevance of the database. There were errors, but also features that had not been clearly communicated. Consequently, all machine learning variant stability predictors have been trained on biased and incorrect data. We obtained a corrected dataset and trained a random forests-based tool, PON-tstab, applicable to variants in any organism. Our results highlight the importance of the benchmark quality, suitability and appropriateness. Predictions are provided for three categories: stability decreasing, increasing and those not affecting stability.

Unearthing new genomic markers of drug response by improved measurement of discriminative power

Background

Oncology drugs are only effective in a small proportion of cancer patients. Our current ability to identify these responsive patients before treatment is still poor in most cases. Thus, there is a pressing need to discover response markers for marketed and research oncology drugs. Screening these drugs against a large panel of cancer cell lines has led to the discovery of new genomic markers of in vitro drug response. However, while the identification of such markers among thousands of candidate drug-gene associations in the data is error-prone, an appraisal of the effectiveness of such detection task is currently lacking.

Methods

Here we present a new non-parametric method to measuring the discriminative power of a drug-gene association. Unlike parametric statistical tests, the adopted non-parametric test has the advantage of not making strong assumptions about the data distorting the identification of genomic markers. Furthermore, we introduce a new benchmark to further validate these markers in vitro using more recent data not used to identify the markers.

Results

The application of this new methodology has led to the identification of 128 new genomic markers distributed across 61% of the analysed drugs, including 5 drugs without previously known markers, which were missed by the MANOVA test initially applied to analyse data from the Genomics of Drug Sensitivity in Cancer consortium.

Conclusions

Discovering markers using more than one statistical test and testing them on independent data is unusual. We found this helpful to discard statistically significant drug-gene associations that were actually spurious correlations. This approach also revealed new, independently validated, in vitro markers of drug response such as Temsirolimus-CDKN2A (resistance) and Gemcitabine-EWS_FLI1 (sensitivity).

Electronic supplementary material

The online version of this article (10.1186/s12920-018-0336-z) contains supplementary material, which is available to authorized users.

Principles and methods of in-silico prioritization of non-coding regulatory variants

Over a decade of genome-wide association, studies have made great strides toward the detection of genes and genetic mechanisms underlying complex traits. However, the majority of associated loci reside in non-coding regions that are functionally uncharacterized in general. Now, the availability of large-scale tissue and cell type-specific transcriptome and epigenome data enables us to elucidate how non-coding genetic variants can affect gene expressions and are associated with phenotypic changes. Here, we provide an overview of this emerging field in human genomics, summarizing available data resources and state-of-the-art analytic methods to facilitate in-silico prioritization of non-coding regulatory mutations. We also highlight the limitations of current approaches and discuss the direction of much-needed future research.

PON-SC - program for identifying steric clashes caused by amino acid substitutions

Background

Amino acid substitutions due to DNA nucleotide replacements are frequently disease-causing because of affecting functionally important sites. If the substituting amino acid does not fit into the protein, it causes structural alterations that are often harmful. Clashes of amino acids cause local or global structural changes. Testing structural compatibility of variations has been difficult due to the lack of a dedicated method that could handle vast amounts of variation data produced by next generation sequencing technologies.

Results

We developed a method, PON-SC, for detecting protein structural clashes due to amino acid substitutions. The method utilizes side chain rotamer library and tests whether any of the common rotamers can be fitted into the protein structure. The tool was tested both with variants that cause and do not cause clashes and found to have accuracy of 0.71 over five test datasets.

Conclusions

We developed a fast method for residue side chain clash detection. The method provides in addition to the prediction also visualization of the variant in three dimensional structure.

Electronic supplementary material

The online version of this article (10.1186/s12859-017-1947-7) contains supplementary material, which is available to authorized users.

Computer-aided classification of sickle cell retinopathy using quantitative features in optical coherence tomography angiography

As a new optical coherence tomography (OCT) imaging modality, there is no standardized quantitative interpretation of OCT angiography (OCTA) characteristics of sickle cell retinopathy (SCR). This study is to demonstrate computer-aided SCR classification using quantitative OCTA features, i.e., blood vessel tortuosity (BVT), blood vessel diameter (BVD), vessel perimeter index (VPI), foveal avascular zone (FAZ) area, FAZ contour irregularity, parafoveal avascular density (PAD). It was observed that combined features show improved classification performance, compared to single feature. Three classifiers, including support vector machine (SVM), k-nearest neighbor (KNN) algorithm, and discriminant analysis, were evaluated. Sensitivity, specificity, and accuracy were quantified to assess the performance of each classifier. For SCR vs. control classification, all three classifiers performed well with an average accuracy of 95% using the six quantitative OCTA features. For mild vs. severe stage retinopathy classification, SVM shows better (97% accuracy) performance, compared to KNN algorithm (95% accuracy) and discriminant analysis (88% accuracy).

Blind prediction of deleterious amino acid variations with SNPs&GO

SNPs&GO is a machine learning method for predicting the association of single amino acid variations (SAVs) to disease, considering protein functional annotation. The method is a binary classifier that implements a support vector machine algorithm to discriminate between disease-related and neutral SAVs. SNPs&GO combines information from protein sequence with functional annotation encoded by gene ontology (GO) terms. Tested in sequence mode on more than 38,000 SAVs from the SwissVar dataset, our method reached 81% overall accuracy and an area under the receiving operating characteristic curve of 0.88 with low false-positive rate. In almost all the editions of the Critical Assessment of Genome Interpretation (CAGI) experiments, SNPs&GO ranked among the most accurate algorithms for predicting the effect of SAVs. In this paper, we summarize the best results obtained by SNPs&GO on disease-related variations of four CAGI challenges relative to the following genes: CHEK2 (CAGI 2010), RAD50 (CAGI 2011), p16-INK (CAGI 2013), and NAGLU (CAGI 2016). Result evaluation provides insights about the accuracy of our algorithm and the relevance of GO terms in annotating the effect of the variants. It also helps to define good practices for the detection of deleterious SAVs.

Selection of effective cocrystals former for dissolution rate improvement of active pharmaceutical ingredients based on lipoaffinity index

New theoretical screening procedure was proposed for appropriate selection of potential cocrystal formers possessing the ability of enhancing dissolution rates of drugs. The procedure relies on the training set comprising 102 positive and 17 negative cases of cocrystals found in the literature. Despite the fact that the only available data were of qualitative character, performed statistical analysis using binary classification allowed to formulate quantitative criterions. Among considered 3679 molecular descriptors the relative value of lipoaffinity index, expressed as the difference between values calculated for active compound and excipient, has been found as the most appropriate measure suited for discrimination of positive and negative cases. Assuming 5% precision, the applied classification criterion led to inclusion of 70% positive cases in the final prediction. Since lipoaffinity index is a molecular descriptor computed using only 2D information about a chemical structure, its estimation is straightforward and computationally inexpensive. The inclusion of an additional criterion quantifying the cocrystallization probability leads to the following conjunction criterions H_mix<-0.18 and ΔLA>3.61, allowing for identification of dissolution rate enhancers. The screening procedure was applied for finding the most promising coformers of such drugs as Iloperidone, Ritonavir, Carbamazepine and Enthenzamide.

PON-P and PON-P2 predictor performance in CAGI challenges: Lessons learned

Computational tools are widely used for ranking and prioritizing variants for characterizing their disease relevance. Since numerous tools have been developed, they have to be properly assessed before being applied. Critical Assessment of Genome Interpretation (CAGI) experiments have significantly contributed toward the assessment of prediction methods for various tasks. Within and outside the CAGI, we have addressed several questions that facilitate development and assessment of variation interpretation tools. These areas include collection and distribution of benchmark datasets, their use for systematic large-scale method assessment, and the development of guidelines for reporting methods and their performance. For us, CAGI has provided a chance to experiment with new ideas, test the application areas of our methods, and network with other prediction method developers. In this article, we discuss our experiences and lessons learned from the various CAGI challenges. We describe our approaches, their performance, and impact of CAGI on our research. Finally, we discuss some of the possibilities that CAGI experiments have opened up and make some suggestions for future experiments.

Predicting Drug-Induced Cholestasis with the Help of Hepatic Transporters-An in Silico Modeling Approach

Cholestasis represents one out of three types of drug induced liver injury (DILI), which comprises a major challenge in drug development. In this study we applied a two-class classification scheme based on k-nearest neighbors in order to predict cholestasis, using a set of 93 two-dimensional (2D) physicochemical descriptors and predictions of selected hepatic transporters' inhibition (BSEP, BCRP, P-gp, OATP1B1, and OATP1B3). In order to assess the potential contribution of transporter inhibition, we compared whether the inclusion of the transporters' inhibition predictions contributes to a significant increase in model performance in comparison to the plain use of the 93 2D physicochemical descriptors. Our findings were in agreement with literature findings, indicating a contribution not only from BSEP inhibition but a rather synergistic effect deriving from the whole set of transporters. The final optimal model was validated via both 10-fold cross validation and external validation. It performs quite satisfactorily resulting in 0.686 ± 0.013 for accuracy and 0.722 ± 0.014 for area under the receiver operating characteristic curve (AUC) for 10-fold cross-validation (mean ± standard deviation from 50 iterations).

Predicting Severity of Disease-Causing Variants

Most diseases, including those of genetic origin, express a continuum of severity. Clinical interventions for numerous diseases are based on the severity of the phenotype. Predicting severity due to genetic variants could facilitate diagnosis and choice of therapy. Although computational predictions have been used as evidence for classifying the disease relevance of genetic variants, special tools for predicting disease severity in large scale are missing. Here, we manually curated a dataset containing variants leading to severe and less severe phenotypes and studied the abilities of variation impact predictors to distinguish between them. We found that these tools cannot separate the two groups of variants. Then, we developed a novel machine-learning-based method, PON-PS (http://structure.bmc.lu.se/PON-PS), for the classification of amino acid substitutions associated with benign, severe, and less severe phenotypes. We tested the method using an independent test dataset and variants in four additional proteins. For distinguishing severe and nonsevere variants, PON-PS showed an accuracy of 61% in the test dataset, which is higher than for existing tolerance prediction methods. PON-PS is the first generic tool developed for this task. The tool can be used together with other evidence for improving diagnosis and prognosis and for prioritization of preventive interventions, clinical monitoring, and molecular tests.

Elucidating the druggable interface of protein-protein interactions using fragment docking and coevolutionary analysis

Protein-protein interactions play a central role in cellular function. Improving the understanding of complex formation has many practical applications, including the rational design of new therapeutic agents and the mechanisms governing signal transduction networks. The generally large, flat, and relatively featureless binding sites of protein complexes pose many challenges for drug design. Fragment docking and direct coupling analysis are used in an integrated computational method to estimate druggable protein-protein interfaces. (i) This method explores the binding of fragment-sized molecular probes on the protein surface using a molecular docking-based screen. (ii) The energetically favorable binding sites of the probes, called hot spots, are spatially clustered to map out candidate binding sites on the protein surface. (iii) A coevolution-based interface interaction score is used to discriminate between different candidate binding sites, yielding potential interfacial targets for therapeutic drug design. This approach is validated for important, well-studied disease-related proteins with known pharmaceutical targets, and also identifies targets that have yet to be studied. Moreover, therapeutic agents are proposed by chemically connecting the fragments that are strongly bound to the hot spots.

Feature-based classification of human transcription factors into hypothetical sub-classes related to regulatory function

Background

Transcription factors are key proteins in the regulation of gene transcription. An important step in this process is the opening of chromatin in order to make genomic regions available for transcription. Data on DNase I hypersensitivity has previously been used to label a subset of transcription factors as Pioneers, Settlers and Migrants to describe their potential role in this process. These labels represent an interesting hypothesis on gene regulation and possibly a useful approach for data analysis, and therefore we wanted to expand the set of labeled transcription factors to include as many known factors as possible. We have used a well-annotated dataset of 1175 transcription factors as input to supervised machine learning methods, using the subset with previously assigned labels as training set. We then used the final classifier to label the additional transcription factors according to their potential role as Pioneers, Settlers and Migrants. The full set of labeled transcription factors was used to investigate associated properties and functions of each class, including an analysis of interaction data for transcription factors based on DNA co-binding and protein-protein interactions. We also used the assigned labels to analyze a previously published set of gene lists associated with a time course experiment on cell differentiation.

Results

The analysis showed that the classification of transcription factors with respect to their potential role in chromatin opening largely was determined by how they bind to DNA. Each subclass of transcription factors was enriched for properties that seemed to characterize the subclass relative to its role in gene regulation, with very general functions for Pioneers, whereas Migrants to a larger extent were associated with specific processes. Further analysis showed that the expanded classification is a useful resource for analyzing other datasets on transcription factors with respect to their potential role in gene regulation. The analysis of transcription factor interaction data showed complementary differences between the subclasses, where transcription factors labeled as Pioneers often interact with other transcription factors through DNA co-binding, whereas Migrants to a larger extent use protein-protein interactions. The analysis of time course data on cell differentiation indicated a shift in the regulatory program associated with Pioneer-like transcription factors during differentiation.

Conclusions

The expanded classification is an interesting resource for analyzing data on gene regulation, as illustrated here on transcription factor interaction data and data from a time course experiment. The potential regulatory function of transcription factors seems largely to be determined by how they bind DNA, but is also influenced by how they interact with each other through cooperativity and protein-protein interactions.

Electronic supplementary material

The online version of this article (doi:10.1186/s12859-016-1349-2) contains supplementary material, which is available to authorized users.

MiRNATIP: a SOM-based miRNA-target interactions predictor

Background

MicroRNAs (miRNAs) are small non-coding RNA sequences with regulatory functions to post-transcriptional level for several biological processes, such as cell disease progression and metastasis. MiRNAs interact with target messenger RNA (mRNA) genes by base pairing. Experimental identification of miRNA target is one of the major challenges in cancer biology because miRNAs can act as tumour suppressors or oncogenes by targeting different type of targets. The use of machine learning methods for the prediction of the target genes is considered a valid support to investigate miRNA functions and to guide related wet-lab experiments. In this paper we propose the miRNA Target Interaction Predictor (miRNATIP) algorithm, a Self-Organizing Map (SOM) based method for the miRNA target prediction. SOM is trained with the seed region of the miRNA sequences and then the mRNA sequences are projected into the SOM lattice in order to find putative interactions with miRNAs. These interactions will be filtered considering the remaining part of the miRNA sequences and estimating the free-energy necessary for duplex stability.

Results

We tested the proposed method by predicting the miRNA target interactions of both the Homo sapiens and the Caenorhbditis elegans species; then, taking into account validated target (positive) and non-target (negative) interactions, we compared our results with other target predictors, namely miRanda, PITA, PicTar, mirSOM, TargetScan and DIANA-microT, in terms of the most used statistical measures. We demonstrate that our method produces the greatest number of predictions with respect to the other ones, exhibiting good results for both species, reaching the for example the highest percentage of sensitivity of 31 and 30.5 %, respectively for Homo sapiens and for C. elegans. All the predicted interaction are freely available at the following url: http://tblab.pa.icar.cnr.it/public/miRNATIP/.

Conclusions

Results state miRNATIP outperforms or is comparable to the other six state-of-the-art methods, in terms of validated target and non-target interactions, respectively.

KinMutRF: a random forest classifier of sequence variants in the human protein kinase superfamily

Background

The association between aberrant signal processing by protein kinases and human diseases such as cancer was established long time ago. However, understanding the link between sequence variants in the protein kinase superfamily and the mechanistic complex traits at the molecular level remains challenging: cells tolerate most genomic alterations and only a minor fraction disrupt molecular function sufficiently and drive disease.

Results

KinMutRF is a novel random-forest method to automatically identify pathogenic variants in human kinases. Twenty six decision trees implemented as a random forest ponder a battery of features that characterize the variants: a) at the gene level, including membership to a Kinbase group and Gene Ontology terms; b) at the PFAM domain level; and c) at the residue level, the types of amino acids involved, changes in biochemical properties, functional annotations from UniProt, Phospho.ELM and FireDB. KinMutRF identifies disease-associated variants satisfactorily (Acc: 0.88, Prec:0.82, Rec:0.75, F-score:0.78, MCC:0.68) when trained and cross-validated with the 3689 human kinase variants from UniProt that have been annotated as neutral or pathogenic. All unclassified variants were excluded from the training set. Furthermore, KinMutRF is discussed with respect to two independent kinase-specific sets of mutations no included in the training and testing, Kin-Driver (643 variants) and Pon-BTK (1495 variants). Moreover, we provide predictions for the 848 protein kinase variants in UniProt that remained unclassified.

A public implementation of KinMutRF, including documentation and examples, is available online (http://kinmut2.bioinfo.cnio.es). The source code for local installation is released under a GPL version 3 license, and can be downloaded from https://github.com/Rbbt-Workflows/KinMut2.

Conclusions

KinMutRF is capable of classifying kinase variation with good performance. Predictions by KinMutRF compare favorably in a benchmark with other state-of-the-art methods (i.e. SIFT, Polyphen-2, MutationAssesor, MutationTaster, LRT, CADD, FATHMM, and VEST). Kinase-specific features rank as the most elucidatory in terms of information gain and are likely the improvement in prediction performance. This advocates for the development of family-specific classifiers able to exploit the discriminatory power of features unique to individual protein families.

Electronic supplementary material

The online version of this article (doi:10.1186/s12864-016-2723-1) contains supplementary material, which is available to authorized users.

Periscope: quantitative prediction of soluble protein expression in the periplasm of Escherichia coli

Periplasmic expression of soluble proteins in Escherichia coli not only offers a much-simplified downstream purification process, but also enhances the probability of obtaining correctly folded and biologically active proteins. Different combinations of signal peptides and target proteins lead to different soluble protein expression levels, ranging from negligible to several grams per litre. Accurate algorithms for rational selection of promising candidates can serve as a powerful tool to complement with current trial-and-error approaches. Accordingly, proteomics studies can be conducted with greater efficiency and cost-effectiveness. Here, we developed a predictor with a two-stage architecture, to predict the real-valued expression level of target protein in the periplasm. The output of the first-stage support vector machine (SVM) classifier determines which second-stage support vector regression (SVR) classifier to be used. When tested on an independent test dataset, the predictor achieved an overall prediction accuracy of 78% and a Pearson's correlation coefficient (PCC) of 0.77. We further illustrate the relative importance of various features with respect to different models. The results indicate that the occurrence of dipeptide glutamine and aspartic acid is the most important feature for the classification model. Finally, we provide access to the implemented predictor through the Periscope webserver, freely accessible at http://lightning.med.monash.edu/periscope/.

PON-Sol: prediction of effects of amino acid substitutions on protein solubility

MOTIVATION

Solubility is one of the fundamental protein properties. It is of great interest because of its relevance to protein expression. Reduced solubility and protein aggregation are also associated with many diseases.

RESULTS

We collected from literature the largest experimentally verified solubility affecting amino acid substitution (AAS) dataset and used it to train a predictor called PON-Sol. The predictor can distinguish both solubility decreasing and increasing variants from those not affecting solubility. PON-Sol has normalized correct prediction ratio of 0.491 on cross-validation and 0.432 for independent test set. The performance of the method was compared both to solubility and aggregation predictors and found to be superior. PON-Sol can be used for the prediction of effects of disease-related substitutions, effects on heterologous recombinant protein expression and enhanced crystallizability. One application is to investigate effects of all possible AASs in a protein to aid protein engineering.

AVAILABILITY AND IMPLEMENTATION

PON-Sol is freely available at http://structure.bmc.lu.se/PON-Sol The training and test data are available at http://structure.bmc.lu.se/VariBench/ponsol.php

CONTACT

mauno.vihinen@med.lu.se

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

PON-mt-tRNA: a multifactorial probability-based method for classification of mitochondrial tRNA variations

Transfer RNAs (tRNAs) are essential for encoding the transcribed genetic information from DNA into proteins. Variations in the human tRNAs are involved in diverse clinical phenotypes. Interestingly, all pathogenic variations in tRNAs are located in mitochondrial tRNAs (mt-tRNAs). Therefore, it is crucial to identify pathogenic variations in mt-tRNAs for disease diagnosis and proper treatment. We collected mt-tRNA variations using a classification based on evidence from several sources and used the data to develop a multifactorial probability-based prediction method, PON-mt-tRNA, for classification of mt-tRNA single nucleotide substitutions. We integrated a machine learning-based predictor and an evidence-based likelihood ratio for pathogenicity using evidence of segregation, biochemistry and histochemistry to predict the posterior probability of pathogenicity of variants. The accuracy and Matthews correlation coefficient (MCC) of PON-mt-tRNA are 1.00 and 0.99, respectively. In the absence of evidence from segregation, biochemistry and histochemistry, PON-mt-tRNA classifies variations based on the machine learning method with an accuracy and MCC of 0.69 and 0.39, respectively. We classified all possible single nucleotide substitutions in all human mt-tRNAs using PON-mt-tRNA. The variations in the loops are more often tolerated compared to the variations in stems. The anticodon loop contains comparatively more predicted pathogenic variations than the other loops. PON-mt-tRNA is available at http://structure.bmc.lu.se/PON-mt-tRNA/.

Investigating the linkage between disease-causing amino acid variants and their effect on protein stability and binding

Single amino acid variations (SAV) occurring in human population result in natural differences between individuals or cause diseases. It is well understood that the molecular effect of SAV can be manifested as changes of the wild type characteristics of the corresponding protein, among which are the protein stability and protein interactions. Typically the effect of SAV on protein stability and interactions was assessed via the changes of the wild type folding and binding free energies. However, in terms of SAV affecting protein functionally and disease susceptibility, one wants to know to what extend the wild type function is perturbed by the SAV. Here it is demonstrated that relative, rather than the absolute, change of the folding and binding free energy serves as a good indicator for SAV association with disease. Using HumVar as a source for disease-causing SAV and experimentally determined free energy changes from ProTherm and SKEMPI databases, correlation coefficients (CC) between the disease index (Pd) and relative folding (Ppr,f) and binding (Ppr,b) probability indexes, respectively, was achieved. The obtained CCs demonstrated the applicability of the proposed approach and it served as good indicator for SAV association with disease.

Evaluation of Second-Level Inference in fMRI Analysis

We investigate the impact of decisions in the second-level (i.e., over subjects) inferential process in functional magnetic resonance imaging on (1) the balance between false positives and false negatives and on (2) the data-analytical stability, both proxies for the reproducibility of results. Second-level analysis based on a mass univariate approach typically consists of 3 phases. First, one proceeds via a general linear model for a test image that consists of pooled information from different subjects. We evaluate models that take into account first-level (within-subjects) variability and models that do not take into account this variability. Second, one proceeds via inference based on parametrical assumptions or via permutation-based inference. Third, we evaluate 3 commonly used procedures to address the multiple testing problem: familywise error rate correction, False Discovery Rate (FDR) correction, and a two-step procedure with minimal cluster size. Based on a simulation study and real data we find that the two-step procedure with minimal cluster size results in most stable results, followed by the familywise error rate correction. The FDR results in most variable results, for both permutation-based inference and parametrical inference. Modeling the subject-specific variability yields a better balance between false positives and false negatives when using parametric inference.

Predicting stabilizing mutations in proteins using Poisson-Boltzmann based models: study of unfolded state ensemble models and development of a successful binary classifier based on residue interaction energies

In many cases the stability of a protein has to be increased to permit its biotechnological use. Rational methods of protein stabilization based on optimizing electrostatic interactions have provided some fine successful predictions. However, the precise calculation of stabilization energies remains challenging, one reason being that the electrostatic effects on the unfolded state are often neglected. We have explored here the feasibility of incorporating Poisson-Boltzmann model electrostatic calculations performed on representations of the unfolded state as large ensembles of geometrically optimized conformations calculated using the ProtSA server. Using a data set of 80 electrostatic mutations experimentally tested in two-state proteins, the predictive performance of several such models has been compared to that of a simple one that considers an unfolded structure of non-interacting residues. The unfolded ensemble models, while showing correlation between the predicted stabilization values and the experimental ones, are worse than the simple model, suggesting that the ensembles do not capture well the energetics of the unfolded state. A more attainable goal is classifying potential mutations as either stabilizing or non-stabilizing, rather than accurately calculating their stabilization energies. To implement a fast classification method that can assist in selecting stabilizing mutations, we have used a much simpler electrostatic model based only on the native structure and have determined its precision using different stabilizing energy thresholds. The binary classifier developed finds 7 true stabilizing mutants out of every 10 proposed candidates and can be used as a robust tool to propose stabilizing mutations.

wKinMut-2: Identification and Interpretation of Pathogenic Variants in Human Protein Kinases

Most genomic alterations are tolerated while only a minor fraction disrupts molecular function sufficiently to drive disease. Protein kinases play a central biological function and the functional consequences of their variants are abundantly characterized. However, this heterogeneous information is often scattered across different sources, which makes the integrative analysis complex and laborious. wKinMut-2 constitutes a solution to facilitate the interpretation of the consequences of human protein kinase variation. Nine methods predict their pathogenicity, including a kinase-specific random forest approach. To understand the biological mechanisms causative of human diseases and cancer, information from pertinent reference knowledge bases and the literature is automatically mined, digested, and homogenized. Variants are visualized in their structural contexts and residues affecting catalytic and drug binding are identified. Known protein-protein interactions are reported. Altogether, this information is intended to assist the generation of new working hypothesis to be corroborated with ulterior experimental work. The wKinMut-2 system, along with a user manual and examples, is freely accessible at http://kinmut2.bioinfo.cnio.es, the code for local installations can be downloaded from https://github.com/Rbbt-Workflows/KinMut2.

Performance of In Silico Tools for the Evaluation of UGT1A1 Missense Variants

Variations in the gene encoding uridine diphosphate glucuronosyltransferase 1A1 (UGT1A1) are particularly important because they have been associated with hyperbilirubinemia in Gilbert's and Crigler-Najjar syndromes as well as with changes in drug metabolism. Several variants associated with these phenotypes are nonsynonymous single-nucleotide polymorphisms (nsSNPs). Bioinformatics approaches have gained increasing importance in predicting the functional significance of these variants. This study was focused on the predictive ability of bioinformatics approaches to determine the pathogenicity of human UGT1A1 nsSNPs, which were previously characterized at the protein level by in vivo and in vitro studies. Using 16 Web algorithms, we evaluated 48 nsSNPs described in the literature and databases. Eight of these algorithms reached or exceeded 90% sensitivity and six presented a Matthews correlation coefficient above 0.46. The best-performing method was MutPred, followed by Sorting Intolerant from Tolerant (SIFT). The prediction measures varied significantly when predictors such us SIFT, polyphen-2, and Prediction of Pathological Mutations on Proteins were run with their native alignment generated by the tool, or with an input alignment that was strictly built with UGT1A1 orthologs and manually curated. Our results showed that the prediction performance of some methods based on sequence conservation analysis can be negatively affected when nsSNPs are positioned at the hypervariable or constant regions of UGT1A1 ortholog sequences.

Classification of Amino Acid Substitutions in Mismatch Repair Proteins Using PON-MMR2

Variations in mismatch repair (MMR) system genes are causative of Lynch syndrome and other cancers. Thousands of variants have been identified in MMR genes, but the clinical relevance is known for only a small proportion. Recently, the InSiGHT group classified 2,360 MMR variants into five classes. One-third of variants, majority of which is nonsynonymous variants, remain to be of uncertain clinical relevance. Computational tools can be used to prioritize variants for disease relevance investigations. Previously, we classified 248 MMR variants as likely pathogenic and likely benign using PON-MMR. We have developed a novel tool, PON-MMR2, which is trained on a larger and more reliable dataset. In performance comparison, PON-MMR2 outperforms both generic tolerance prediction methods as well as methods optimized for MMR variants. It achieves accuracy and MCC of 0.89 and 0.78, respectively, in cross-validation and 0.86 and 0.69, respectively, on an independent test dataset. We classified 354 class 3 variants in InSiGHT database as well as all possible amino acid substitutions in four MMR proteins. Likely harmful variants mainly appear in the protein core, whereas likely benign variants are on the surface. PON-MMR2 is a highly reliable tool to prioritize variants for functional analysis. It is freely available at http://structure.bmc.lu.se/PON-MMR2/.

Relationships Between Pharmacovigilance, Molecular, Structural, and Pathway Data: Revealing Mechanisms for Immune-Mediated Drug-Induced Liver Injury

Immune-mediated drug-induced liver injury (IMDILI) can be devastating, irreversible, and fatal in the absence of successful transplantation surgery. We present a novel approach that combines the methods of pharmacoepidemiology with in silico molecular modeling to identify specific features in toxic ligands that are associated with clinical features of IMDILI. Specifically, from pharmacovigilance data multivariate logistic regression identified 18 drugs associated with IMDILI (P < 0.00015). Eleven of these drugs, along with their known and proposed metabolites, constituted a training set used to develop a four-point pharmacophore model (sensitivity 75%; specificity 85%). Subsequently, this information was combined with information from immune-pathway reviews and genetic-association studies and complemented with ligand-protein docking simulations to support a hypothesis implicating two putative targets within separate, possibly interacting, immune-system pathways: the major histocompatibility complex within the adaptive immune system and Toll-like receptors (TLRs), in particular TLR-7, which represent pattern recognition receptors of the innate immune system.

Predicting Metabolic Syndrome Using the Random Forest Method

Aims. This study proposes a computational method for determining the prevalence of metabolic syndrome (MS) and to predict its occurrence using the National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP III) criteria. The Random Forest (RF) method is also applied to identify significant health parameters. Materials and Methods. We used data from 5,646 adults aged between 18–78 years residing in Bangkok who had received an annual health check-up in 2008. MS was identified using the NCEP ATP III criteria. The RF method was applied to predict the occurrence of MS and to identify important health parameters surrounding this disorder. Results. The overall prevalence of MS was 23.70% (34.32% for males and 17.74% for females). RF accuracy for predicting MS in an adult Thai population was 98.11%. Further, based on RF, triglyceride levels were the most important health parameter associated with MS. Conclusion. RF was shown to predict MS in an adult Thai population with an accuracy >98% and triglyceride levels were identified as the most informative variable associated with MS. Therefore, using RF to predict MS may be potentially beneficial in identifying MS status for preventing the development of diabetes mellitus and cardiovascular diseases.

Assessing models for genetic prediction of complex traits: a comparison of visualization and quantitative methods

Background

In silico models have recently been created in order to predict which genetic variants are more likely to contribute to the risk of a complex trait given their functional characteristics. However, there has been no comprehensive review as to which type of predictive accuracy measures and data visualization techniques are most useful for assessing these models.

Methods

We assessed the performance of the models for predicting risk using various methodologies, some of which include: receiver operating characteristic (ROC) curves, histograms of classification probability, and the novel use of the quantile-quantile plot. These measures have variable interpretability depending on factors such as whether the dataset is balanced in terms of numbers of genetic variants classified as risk variants versus those that are not.

Results

We conclude that the area under the curve (AUC) is a suitable starting place, and for models with similar AUCs, violin plots are particularly useful for examining the distribution of the risk scores.

Electronic supplementary material

The online version of this article (doi:10.1186/s12864-015-1616-z) contains supplementary material, which is available to authorized users.

Assessment of the predictive accuracy of five in silico prediction tools, alone or in combination, and two metaservers to classify long QT syndrome gene mutations

Background

Long QT syndrome (LQTS) is an autosomal dominant condition predisposing to sudden death from malignant arrhythmia. Genetic testing identifies many missense single nucleotide variants of uncertain pathogenicity. Establishing genetic pathogenicity is an essential prerequisite to family cascade screening. Many laboratories use in silico prediction tools, either alone or in combination, or metaservers, in order to predict pathogenicity; however, their accuracy in the context of LQTS is unknown. We evaluated the accuracy of five in silico programs and two metaservers in the analysis of LQTS 1–3 gene variants.

Methods

The in silico tools SIFT, PolyPhen-2, PROVEAN, SNPs&GO and SNAP, either alone or in all possible combinations, and the metaservers Meta-SNP and PredictSNP, were tested on 312 KCNQ1, KCNH2 and SCN5A gene variants that have previously been characterised by either in vitro or co-segregation studies as either “pathogenic” (283) or “benign” (29). The accuracy, sensitivity, specificity and Matthews Correlation Coefficient (MCC) were calculated to determine the best combination of in silico tools for each LQTS gene, and when all genes are combined.

Results

The best combination of in silico tools for KCNQ1 is PROVEAN, SNPs&GO and SIFT (accuracy 92.7%, sensitivity 93.1%, specificity 100% and MCC 0.70). The best combination of in silico tools for KCNH2 is SIFT and PROVEAN or PROVEAN, SNPs&GO and SIFT. Both combinations have the same scores for accuracy (91.1%), sensitivity (91.5%), specificity (87.5%) and MCC (0.62). In the case of SCN5A, SNAP and PROVEAN provided the best combination (accuracy 81.4%, sensitivity 86.9%, specificity 50.0%, and MCC 0.32). When all three LQT genes are combined, SIFT, PROVEAN and SNAP is the combination with the best performance (accuracy 82.7%, sensitivity 83.0%, specificity 80.0%, and MCC 0.44). Both metaservers performed better than the single in silico tools; however, they did not perform better than the best performing combination of in silico tools.

Conclusions

The combination of in silico tools with the best performance is gene-dependent. The in silico tools reported here may have some value in assessing variants in the KCNQ1 and KCNH2 genes, but caution should be taken when the analysis is applied to SCN5A gene variants.

Electronic supplementary material

The online version of this article (doi:10.1186/s12881-015-0176-z) contains supplementary material, which is available to authorized users.

Assessment of the predictive accuracy of five in silico prediction tools, alone or in combination, and two metaservers to classify long QT syndrome gene mutations

BACKGROUND

Long QT syndrome (LQTS) is an autosomal dominant condition predisposing to sudden death from malignant arrhythmia. Genetic testing identifies many missense single nucleotide variants of uncertain pathogenicity. Establishing genetic pathogenicity is an essential prerequisite to family cascade screening. Many laboratories use in silico prediction tools, either alone or in combination, or metaservers, in order to predict pathogenicity; however, their accuracy in the context of LQTS is unknown. We evaluated the accuracy of five in silico programs and two metaservers in the analysis of LQTS 1-3 gene variants.

METHODS

The in silico tools SIFT, PolyPhen-2, PROVEAN, SNPs&GO and SNAP, either alone or in all possible combinations, and the metaservers Meta-SNP and PredictSNP, were tested on 312 KCNQ1, KCNH2 and SCN5A gene variants that have previously been characterised by either in vitro or co-segregation studies as either "pathogenic" (283) or "benign" (29). The accuracy, sensitivity, specificity and Matthews Correlation Coefficient (MCC) were calculated to determine the best combination of in silico tools for each LQTS gene, and when all genes are combined.

RESULTS

The best combination of in silico tools for KCNQ1 is PROVEAN, SNPs&GO and SIFT (accuracy 92.7%, sensitivity 93.1%, specificity 100% and MCC 0.70). The best combination of in silico tools for KCNH2 is SIFT and PROVEAN or PROVEAN, SNPs&GO and SIFT. Both combinations have the same scores for accuracy (91.1%), sensitivity (91.5%), specificity (87.5%) and MCC (0.62). In the case of SCN5A, SNAP and PROVEAN provided the best combination (accuracy 81.4%, sensitivity 86.9%, specificity 50.0%, and MCC 0.32). When all three LQT genes are combined, SIFT, PROVEAN and SNAP is the combination with the best performance (accuracy 82.7%, sensitivity 83.0%, specificity 80.0%, and MCC 0.44). Both metaservers performed better than the single in silico tools; however, they did not perform better than the best performing combination of in silico tools.

CONCLUSIONS

The combination of in silico tools with the best performance is gene-dependent. The in silico tools reported here may have some value in assessing variants in the KCNQ1 and KCNH2 genes, but caution should be taken when the analysis is applied to SCN5A gene variants.

Characterization of all possible single-nucleotide change caused amino acid substitutions in the kinase domain of Bruton tyrosine kinase

Knowledge about features distinguishing deleterious and neutral variations is crucial for interpretation of novel variants. Bruton tyrosine kinase (BTK) contains the highest number of unique disease-causing variations among the human protein kinases, still it is just 10% of all the possible single-nucleotide substitution-caused amino acid variations (SNAVs). In the BTK kinase domain (BTK-KD) can appear altogether 1,495 SNAVs. We investigated them all with bioinformatic and protein structure analysis methods. Most disease-causing variations affect conserved and buried residues disturbing protein stability. Minority of exposed residues is conserved, but strongly tied to pathogenicity. Sixty-seven percent of variations are predicted to be harmful. In 39% of the residues, all the variants are likely harmful, whereas in 10% of sites, all the substitutions are tolerated. Results indicate the importance of the entire kinase domain, involvement in numerous interactions, and intricate functional regulation by conformational change. These results can be extended to other protein kinases and organisms.

Multi-Step Protocol for Automatic Evaluation of Docking Results Based on Machine Learning Methods--A Case Study of Serotonin Receptors 5-HT(6) and 5-HT(7)

Molecular docking, despite its undeniable usefulness in computer-aided drug design protocols and the increasing sophistication of tools used in the prediction of ligand-protein interaction energies, is still connected with a problem of effective results analysis. In this study, a novel protocol for the automatic evaluation of numerous docking results is presented, being a combination of Structural Interaction Fingerprints and Spectrophores descriptors, machine-learning techniques, and multi-step results analysis. Such an approach takes into consideration the performance of a particular learning algorithm (five machine learning methods were applied), the performance of the docking algorithm itself, the variety of conformations returned from the docking experiment, and the receptor structure (homology models were constructed on five different templates). Evaluation using compounds active toward 5-HT6 and 5-HT7 receptors, as well as additional analysis carried out for beta-2 adrenergic receptor ligands, proved that the methodology is a viable tool for supporting virtual screening protocols, enabling proper discrimination between active and inactive compounds.

The evaluation of tools used to predict the impact of missense variants is hindered by two types of circularity

Prioritizing missense variants for further experimental investigation is a key challenge in current sequencing studies for exploring complex and Mendelian diseases. A large number of in silico tools have been employed for the task of pathogenicity prediction, including PolyPhen-2, SIFT, FatHMM, MutationTaster-2, MutationAssessor, Combined Annotation Dependent Depletion, LRT, phyloP, and GERP++, as well as optimized methods of combining tool scores, such as Condel and Logit. Due to the wealth of these methods, an important practical question to answer is which of these tools generalize best, that is, correctly predict the pathogenic character of new variants. We here demonstrate in a study of 10 tools on five datasets that such a comparative evaluation of these tools is hindered by two types of circularity: they arise due to (1) the same variants or (2) different variants from the same protein occurring both in the datasets used for training and for evaluation of these tools, which may lead to overly optimistic results. We show that comparative evaluations of predictors that do not address these types of circularity may erroneously conclude that circularity confounded tools are most accurate among all tools, and may even outperform optimized combinations of tools.

Prediction of aromatase inhibitory activity using the efficient linear method (ELM)

Aromatase inhibition is an effective treatment strategy for breast cancer. Currently, several in silico methods have been developed for the prediction of aromatase inhibitors (AIs) using artificial neural network (ANN) or support vector machine (SVM). In spite of this, there are ample opportunities for further improvements by developing a simple and interpretable quantitative structure-activity relationship (QSAR) method. Herein, an efficient linear method (ELM) is proposed for constructing a highly predictive QSAR model containing a spontaneous feature importance estimator. Briefly, ELM is a linear-based model with optimal parameters derived from genetic algorithm. Results showed that the simple ELM method displayed robust performance with 10-fold cross-validation MCC values of 0.64 and 0.56 for steroidal and non-steroidal AIs, respectively. Comparative analyses with other machine learning methods (i.e. ANN, SVM and decision tree) were also performed. A thorough analysis of informative molecular descriptors for both steroidal and non-steroidal AIs provided insights into the mechanism of action of compounds. Our findings suggest that the shape and polarizability of compounds may govern the inhibitory activity of both steroidal and non-steroidal types whereas the terminal primary C(sp3) functional group and electronegativity may be required for non-steroidal AIs. The R code of the ELM method is available at http://dx.doi.org/10.6084/m9.figshare.1274030.

Problems with the nested granularity of feature domains in bioinformatics: the eXtasy case

Background

Data from biomedical domains often have an inherit hierarchical structure. As this structure is usually implicit, its existence can be overlooked by practitioners interested in constructing and evaluating predictive models from such data. Ignoring these constructs leads to potentially problematic and the routinely unrecognized bias in the models and results. In this work, we discuss this bias in detail and propose a simple, sampling-based solution for it. Next, we explore its sources and extent on synthetic data. Finally, we demonstrate how the state-of-the-art variant prioritization framework, eXtasy, benefits from using the described approach in its Random forest-based core classification model.

Results and conclusions

The conducted simulations clearly indicate that the heterogeneous granularity of feature domains poses significant problems for both the standard Random forest classifier and a modification that relies on stratified bootstrapping. Conversely, using the proposed sampling scheme when training the classifier mitigates the described bias. Furthermore, when applied to the eXtasy data under a realistic class distribution scenario, a Random forest learned using the proposed sampling scheme displays much better precision that its standard version, without degrading recall. Moreover, the largest performance gains are achieved in the most important part of the operating range: the top of prioritized gene list.

PON-P2: prediction method for fast and reliable identification of harmful variants

More reliable and faster prediction methods are needed to interpret enormous amounts of data generated by sequencing and genome projects. We have developed a new computational tool, PON-P2, for classification of amino acid substitutions in human proteins. The method is a machine learning-based classifier and groups the variants into pathogenic, neutral and unknown classes, on the basis of random forest probability score. PON-P2 is trained using pathogenic and neutral variants obtained from VariBench, a database for benchmark variation datasets. PON-P2 utilizes information about evolutionary conservation of sequences, physical and biochemical properties of amino acids, GO annotations and if available, functional annotations of variation sites. Extensive feature selection was performed to identify 8 informative features among altogether 622 features. PON-P2 consistently showed superior performance in comparison to existing state-of-the-art tools. In 10-fold cross-validation test, its accuracy and MCC are 0.90 and 0.80, respectively, and in the independent test, they are 0.86 and 0.71, respectively. The coverage of PON-P2 is 61.7% in the 10-fold cross-validation and 62.1% in the test dataset. PON-P2 is a powerful tool for screening harmful variants and for ranking and prioritizing experimental characterization. It is very fast making it capable of analyzing large variant datasets. PON-P2 is freely available at http://structure.bmc.lu.se/PON-P2/.

A method to distinguish between lysine acetylation and lysine ubiquitination with feature selection and analysis

Lysine acetylation and ubiquitination are two primary post-translational modifications (PTMs) in most eukaryotic proteins. Lysine residues are targets for both types of PTMs, resulting in different cellular roles. With the increasing availability of protein sequences and PTM data, it is challenging to distinguish the two types of PTMs on lysine residues. Experimental approaches are often laborious and time consuming. There is an urgent need for computational tools to distinguish between lysine acetylation and ubiquitination. In this study, we developed a novel method, called DAUFSA (distinguish between lysine acetylation and lysine ubiquitination with feature selection and analysis), to discriminate ubiquitinated and acetylated lysine residues. The method incorporated several types of features: PSSM (position-specific scoring matrix) conservation scores, amino acid factors, secondary structures, solvent accessibilities, and disorder scores. By using the mRMR (maximum relevance minimum redundancy) method and the IFS (incremental feature selection) method, an optimal feature set containing 290 features was selected from all incorporated features. A dagging-based classifier constructed by the optimal features achieved a classification accuracy of 69.53%, with an MCC of .3853. An optimal feature set analysis showed that the PSSM conservation score features and the amino acid factor features were the most important attributes, suggesting differences between acetylation and ubiquitination. Our study results also supported previous findings that different motifs were employed by acetylation and ubiquitination. The feature differences between the two modifications revealed in this study are worthy of experimental validation and further investigation.

Comparison and integration of deleteriousness prediction methods for nonsynonymous SNVs in whole exome sequencing studies

Accurate deleteriousness prediction for nonsynonymous variants is crucial for distinguishing pathogenic mutations from background polymorphisms in whole exome sequencing (WES) studies. Although many deleteriousness prediction methods have been developed, their prediction results are sometimes inconsistent with each other and their relative merits are still unclear in practical applications. To address these issues, we comprehensively evaluated the predictive performance of 18 current deleteriousness-scoring methods, including 11 function prediction scores (PolyPhen-2, SIFT, MutationTaster, Mutation Assessor, FATHMM, LRT, PANTHER, PhD-SNP, SNAP, SNPs&GO and MutPred), 3 conservation scores (GERP++, SiPhy and PhyloP) and 4 ensemble scores (CADD, PON-P, KGGSeq and CONDEL). We found that FATHMM and KGGSeq had the highest discriminative power among independent scores and ensemble scores, respectively. Moreover, to ensure unbiased performance evaluation of these prediction scores, we manually collected three distinct testing datasets, on which no current prediction scores were tuned. In addition, we developed two new ensemble scores that integrate nine independent scores and allele frequency. Our scores achieved the highest discriminative power compared with all the deleteriousness prediction scores tested and showed low false-positive prediction rate for benign yet rare nonsynonymous variants, which demonstrated the value of combining information from multiple orthologous approaches. Finally, to facilitate variant prioritization in WES studies, we have pre-computed our ensemble scores for 87 347 044 possible variants in the whole-exome and made them publicly available through the ANNOVAR software and the dbNSFP database.

Molecular damage in Fabry disease: characterization and prediction of alpha-galactosidase A pathological mutations

Loss-of-function mutations of the enzyme alpha-galactosidase A (GLA) causes Fabry disease (FD), that is a rare and potentially fatal disease. Identification of these pathological mutations by sequencing is important because it allows an early treatment of the disease. However, before taking any treatment decision, if the mutation identified is unknown, we first need to establish if it is pathological or not. General bioinformatic tools (PolyPhen-2, SIFT, Condel, etc.) can be used for this purpose, but their performance is still limited. Here we present a new tool, specifically derived for the assessment of GLA mutations. We first compared mutations of this enzyme known to cause FD with neutral sequence variants, using several structure and sequence properties. Then, we used these properties to develop a family of prediction methods adapted to different quality requirements. Trained and tested on a set of known Fabry mutations, our methods have a performance (Matthews correlation: 0.56-0.72) comparable or better than that of the more complex method, Polyphen-2 (Matthews correlation: 0.61), and better than those of SIFT (Matthews correl.: 0.54) and Condel (Matthews correl.: 0.51). This result is validated in an independent set of 65 pathological mutations, for which our method displayed the best success rate (91.0%, 87.7%, and 73.8%, for our method, PolyPhen-2 and SIFT, respectively). These data confirmed that our specific approach can effectively contribute to the identification of pathological mutations in GLA, and therefore enhance the use of sequence information in the identification of undiagnosed Fabry patients.

SuRFing the genomics wave: an R package for prioritising SNPs by functionality

Identifying functional non-coding variants is one of the greatest unmet challenges in genetics. To help address this, we introduce an R package, SuRFR, which integrates functional annotation and prior biological knowledge to prioritise candidate functional variants. SuRFR is publicly available, modular, flexible, fast, and simple to use. We demonstrate that SuRFR performs with high sensitivity and specificity and provide a widely applicable and scalable benchmarking dataset for model training and validation.

Website: http://www.cgem.ed.ac.uk/resources/

Discriminating between lysine sumoylation and lysine acetylation using mRMR feature selection and analysis

Post-translational modifications (PTMs) are crucial steps in protein synthesis and are important factors contributing to protein diversity. PTMs play important roles in the regulation of gene expression, protein stability and metabolism. Lysine residues in protein sequences have been found to be targeted for both types of PTMs: sumoylations and acetylations; however, each PTM has a different cellular role. As experimental approaches are often laborious and time consuming, it is challenging to distinguish the two types of PTMs on lysine residues using computational methods. In this study, we developed a method to discriminate between sumoylated lysine residues and acetylated residues. The method incorporated several features: PSSM conservation scores, amino acid factors, secondary structures, solvent accessibilities and disorder scores. By using the mRMR (Maximum Relevance Minimum Redundancy) method and the IFS (Incremental Feature Selection) method, an optimal feature set was selected from all of the incorporated features, with which the classifier achieved 92.14% accuracy with an MCC value of 0.7322. Analysis of the optimal feature set revealed some differences between acetylation and sumoylation. The results from our study also supported the previous finding that there exist different consensus motifs for the two types of PTMs. The results could suggest possible dominant factors governing the acetylation and sumoylation of lysine residues, shedding some light on the modification dynamics and molecular mechanisms of the two types of PTMs, and provide guidelines for experimental validations.

Majority vote and other problems when using computational tools

Computational tools are essential for most of our research. To use these tools, one needs to know how they work. Problems in application of computational methods to variation analysis can appear at several stages and affect, for example, the interpretation of results. Such cases are discussed along with suggestions how to avoid them. The applications include incomplete reporting of methods, especially about the use of prediction tools; method selection on unscientific grounds and without consulting independent method performance assessments; extending application area of methods outside their intended purpose; use of the same data several times for obtaining majority vote; and filtering of datasets so that variants of interest are excluded. All these issues can be avoided by discontinuing the use software tools as black boxes.

Medical genomics: The intricate path from genetic variant identification to clinical interpretation

The field of medical genomics involves translating high throughput genetic methods to the clinic, in order to improve diagnostic efficiency and treatment decision making. Technical questions related to sample enrichment, sequencing methodologies and variant identification and calling algorithms, still need careful investigation in order to validate the analytical step of next generation sequencing techniques for clinical applications. However, the main foreseeable challenge will be interpreting the clinical significance of the variants observed in a given patient, as well as their significance for family members and for other patients.

Every step in the variant interpretation process has limitations and difficulties, and its quote of contribution to false positive and false negative results. There is no single piece of evidence enough on its own to make firm conclusions on the pathogenicity and disease causality of a given variant.

A plethora of automated analysis software tools is being developed that will enhance efficiency and accuracy. However a risk of misinterpretation could derive from biased biorepository content, facilitated by annotation of variant functional consequences using previous datasets stored in the same or linked repositories. In order to improve variant interpretation and avoid an exponential accumulation of confounding noise in the medical literature, the use of terms in a standard way should be sought and requested when reporting genetic variants and their consequences. Generally, stepwise and linear interpretation processes are likely to overrate some pieces of evidence while underscoring others. Algorithms are needed that allow a multidimensional, parallel analysis of diverse lines of evidence to be carried out by expert teams for specific genes, cellular pathways or disorders.

DEFLATE compression algorithm corrects for overestimation of phylogenetic diversity by Grantham approach to single-nucleotide polymorphism classification

Improvements in speed and cost of genome sequencing are resulting in increasing numbers of novel non-synonymous single nucleotide polymorphisms (nsSNPs) in genes known to be associated with disease. The large number of nsSNPs makes laboratory-based classification infeasible and familial co-segregation with disease is not always possible. In-silico methods for classification or triage are thus utilised. A popular tool based on multiple-species sequence alignments (MSAs) and work by Grantham, Align-GVGD, has been shown to underestimate deleterious effects, particularly as sequence numbers increase. We utilised the DEFLATE compression algorithm to account for expected variation across a number of species. With the adjusted Grantham measure we derived a means of quantitatively clustering known neutral and deleterious nsSNPs from the same gene; this was then used to assign novel variants to the most appropriate cluster as a means of binary classification. Scaling of clusters allows for inter-gene comparison of variants through a single pathogenicity score. The approach improves upon the classification accuracy of Align-GVGD while correcting for sensitivity to large MSAs. Open-source code and a web server are made available at https://github.com/aschlosberg/CompressGV.

Usefulness of the heart-rate variability complex for predicting cardiac mortality after acute myocardial infarction

Background

Previous studies indicate that decreased heart-rate variability (HRV) is related to the risk of death in patients after acute myocardial infarction (AMI). However, the conventional indices of HRV have poor predictive value for mortality. Our aim was to develop novel predictive models based on support vector machine (SVM) to study the integrated features of HRV for improving risk stratification after AMI.

Methods

A series of heart-rate dynamic parameters from 208 patients were analyzed after a mean follow-up time of 28 months. Patient electrocardiographic data were classified as either survivals or cardiac deaths. SVM models were established based on different combinations of heart-rate dynamic variables and compared to left ventricular ejection fraction (LVEF), standard deviation of normal-to-normal intervals (SDNN) and deceleration capacity (DC) of heart rate. We tested the accuracy of predictors by assessing the area under the receiver-operator characteristics curve (AUC).

Results

We evaluated a SVM algorithm that integrated various electrocardiographic features based on three models: (A) HRV complex; (B) 6 dimension vector; and (C) 8 dimension vector. Mean AUC of HRV complex was 0.8902, 0.8880 for 6 dimension vector and 0.8579 for 8 dimension vector, compared with 0.7424 for LVEF, 0.7932 for SDNN and 0.7399 for DC.

Conclusions

HRV complex yielded the largest AUC and is the best classifier for predicting cardiac death after AMI.

MutPred Splice: machine learning-based prediction of exonic variants that disrupt splicing

We have developed a novel machine-learning approach, MutPred Splice, for the identification of coding region substitutions that disrupt pre-mRNA splicing. Applying MutPred Splice to human disease-causing exonic mutations suggests that 16% of mutations causing inherited disease and 10 to 14% of somatic mutations in cancer may disrupt pre-mRNA splicing. For inherited disease, the main mechanism responsible for the splicing defect is splice site loss, whereas for cancer the predominant mechanism of splicing disruption is predicted to be exon skipping via loss of exonic splicing enhancers or gain of exonic splicing silencer elements. MutPred Splice is available at http://mutdb.org/mutpredsplice.