PredictSNP: robust and accurate consensus classifier for prediction of disease-related mutations.

Bendl J, Stourac J, Salanda O, Pavelka A, Wieben ED, Zendulka J, Brezovsky J, Damborsky J. · 2014

Read PredictSNP: robust and accurate consensus classifier for prediction of disease-related mutations. by Bendl J, Stourac J, Salanda O, Pavelka A, Wieben ED, Zendulka J, Brezovsky J, Damborsky J. (2014) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc3894168.txt.

Introduction

Jaroslav Bendl and co-authors built a mutation predictor described in PLoS Computational Biology, with the record dated 2014. Single letter DNA changes in coding regions cause many genetic diseases. Computer tools forecast which changes harm a protein, yet their reported scores are inflated by overlap between training and test data.

The team built clean test sets with all such overlap stripped out, holding tens of thousands of mutations. Eight known tools were judged fairly on that ground. The six best were merged into one consensus tool, PredictSNP, which beat each part alone and returned a call for every mutation. The work gives clinics and labs a more honest ranking of which variants merit lab follow-up.

Catalog record

CREATORBendl J, Stourac J, Salanda O, Pavelka A, Wieben ED, Zendulka J, Brezovsky J, Damborsky J.
YEAR2014
COLLECTIONPAPERS
FORMATTXT
FILE SIZE47 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC3894168

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

PredictSNP: Robust and Accurate Consensus Classifier for Prediction of Disease-Related Mutations By Jaroslav Bendl, Jan Stourac, Ondrej Salanda, Antonin Pavelka, Eric D. Wieben, Jaroslav Zendulka, Jan Brezovsky, Jiri Damborsky, Paul P. Gardner Published in: PLoS Computational Biology DOI: 10.1371/journal.pcbi.1003440 ABSTRACT Single nucleotide variants represent a prevalent form of genetic variation. Mutations in the coding regions are frequently associated with the development of various genetic diseases. Computational tools for the prediction of the effects of mutations on protein function are very important for analysis of single nucleotide variants and their prioritization for experimental characterization. Many computational tools are already widely employed for this purpose. Unfortunately, their comparison and further improvement is hindered by large overlaps between the training datasets and benchmark datasets, which lead to biased and overly optimistic reported performances. In this study, we have constructed three independent datasets by removing all duplicities, inconsistencies and mutations previously used in the training of evaluated tools. The benchmark dataset containing over 43,000 mutations was employed for the unbiased evaluation of eight established prediction tools: MAPP, nsSNPAnalyzer, PANTHER, PhD-SNP, PolyPhen-1, PolyPhen-2, SIFT and SNAP. The six best performing tools were combined into a consensus classifier PredictSNP, resulting into significantly improved prediction performance, and at the same time returned results for all mutations, confirming that consensus prediction represents an accurate and robust alternative to the predictions delivered by individual tools. A user-friendly web interface enables easy access to all eight prediction tools, the consensus classifier PredictSNP and annotations from the Protein Mutant Database and the UniProt database. The web server and the datasets are freely available to the academic community at http://loschmidt.chemi.muni.cz/predictsnp. "This is a PLOS Computational Biology Software Article" INTRODUCTION The single nucleotide variants (SNVs) are the most frequent type of genetic variation in humans, responsible for almost 90% of known sequence differences [1], [2]. Although many of these changes are neutral, some variants do affect gene expression or the function of the translated proteins [3], [4]. Such SNVs often have dramatic phenotypic consequences leading to the development of various diseases [5]. Approximately half of the known disease-related mutations stems from non-synonymous SNVs, manifested as amino acid mutations [6], [7]. Although it is extremely important to uncover the links between SNVs and associated diseases, it is difficult to distinguish pathogenic substitutions from those that are functionally neutral by any experimental assay due to rapid growth of the number of known SNVs [8], [9]. Therefore, computational prediction tools became valuable for the initial analysis of SNVs and their prioritization for experimental characterization. There are many computational tools for prediction of the effects of amino acid substitution on protein function, e.g., MutPred [10], nsSNPAnalyzer [11], PolyPhen-1 (PPH-1) [12], PolyPhen-2 (PPH-2) [13], SNAP [14], MAPP [15], PANTHER [16], PhD-SNP [17], SIFT [18] and SNPs&GO [19]. Most of these tools are designed to predict whether a particular substitution is neutral or deleterious, based on various parameters derived from the evolutionary, physico-chemical or structural characteristics [20], [21]. These tools mainly employ machine learning methods to derive their decision rules based on a training datasets of annotated mutations. Since overlaps between the training datasets of evaluated tools and the benchmark datasets have been shown to result in illegitimately high performance estimates of such tools [22], [23], it is of the utmost importance to carry out any comparisons on fully independent datasets [24], [25]. Variability of the training datasets utilized by individual prediction tools coupled with the public inaccessibility of these datasets represent a serious obstacle to unbiased comparison of predictive power of the tools [21]. Since individual prediction tools have been developed using different: (i) training datasets, (ii) machine learning methods and (iii) underlying principles, it is generally believed that some of them can be combined to give a single consensus prediction with improved accuracy [26]. Recent examples of consensus tools are CONDEL [27], PON-P [28] and Meta-SNP [29], all of which reported improved performance over individual integrated tools. In this paper, we constructed three fully independent datasets, one benchmark and two testing datasets, suitable for assessment of the performance of eight selected prediction tools in an unbiased manner. We then combined six best performing methods into a consensus classifier PredictSNP. The developed consensus of these prediction tools provided significant improvement in prediction performance over the individual tools and also over three previously developed consensus classifiers. Finally, we developed a web interface to allow an easy access to all eight prediction tools and consensus PredictSNP. Predictions from the computational tools are supplemented by experimental annotations from Protein Mutant Database [30] and UniProt database [31]. DESIGN AND IMPLEMENTATION Prediction Tools Initially, eight selected prediction tools were chosen for the evaluation (Table 1). The tools had to satisfy following criteria: (i) to allow submission of user-defined sequence, (ii) to be available as a stand-alone application to allow large-scale evaluations and provide stable service, and (iii) not to require a protein structure for the prediction since structural information is available only for a small portion of known sequences. Four selected tools – nsSNPAnalyzer, PhD-SNP, PPH-2, SNAP – each use a decision model trained by various machine-learning methods. Out of the remaining tools, SIFT and PANTHER use solely evolutionary information, while MAPP also considers the differences in physicochemical properties between wild-type and mutant amino acids. Finally, PPH-1 employs an expert set of empirical rules for the classification [32]. All selected tools were installed on local server and run using their default settings with the following modifications. A pipeline developed in the framework of the HotSpot Wizard server [33] was used for construction of multiple sequence alignments and a phylogenetic tree for MAPP. In short, a BLAST sequence search [34] against the non-redundant database at NCBI [35] is performed to gather protein sequences similar to the query. Sequences are clustered by CD-HIT 4.6 [36] and representatives of the clusters aligned using MUSCLE 3.8 [37]. Then, a phylogenetic tree is calculated by Rate4Site 2.01 [38]. PPH-2 offers a choice from two machine-learning models trained by Naïve Bayes on different datasets – HumDiv [13] and HumVar [13]. Only the HumDiv-trained model was employed in this study since HumDiv dataset was constructed using additional criteria to reduce the number of possibly erroneous annotations [13]. SIFT can employ two sequence databases for homologues identification: the non-redundant database at NCBI or UniProt [31], the former being used in this study. Since the SIFT algorithm is unable to process very long sequences, we automatically truncated these sequences into a fragments of 700 amino acids with analyzed mutation located in theirs centers as recommended in the user manual. Finally, median sequence conservation of SIFT method was set to three according to the recommended range. [Table: Principles and training datasets of eight evaluated tools.] Datasets Benchmark dataset The benchmark dataset used for the evaluation of the selected prediction tools and training of consensus classifier PredictSNP was compiled from five different sources. The first four constitute training datasets of the tools, which were not selected for evaluation since they did not meet the selection criteria defined in the previous section. These sources were following: SNPs&GO [19] dataset of 58,057 mutations compiled from SwissProt, the MutPred [10] dataset of 65,654 mutations compiled from SwissProt and HGMD [39] and the PON-P [28] training dataset of

Read the complete text