An integrative approach to predicting the functional effects of non-coding and coding sequence variation.

Shihab HA, Rogers MF, Gough J, Mort M, Cooper DN, Day IN, Gaunt TR, Campbell C. · 2015

Read An integrative approach to predicting the functional effects of non-coding and coding sequence variation. by Shihab HA, Rogers MF, Gough J, Mort M, Cooper DN, Day IN, Gaunt TR, Campbell C. (2015) 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-pmc4426838.txt.

Introduction

Hashem Shihab and seven co-authors presented a prediction tool in Bioinformatics in 2015 for judging what a single-letter DNA change does. Sequencing now finds huge numbers of such variants, in both gene-coding and non-coding regions, and many may affect disease. Older tools often handled only one of those two classes well.

Their method, FATHMM-MKL, combines many kinds of genome notes and learns how much weight each kind deserves. In tests it beat two leading tools, CADD and GWAVA, on non-coding variants and matched the best tools on coding ones. It also gives a confidence score to rank its calls. A free web server lets other researchers submit variants and get predictions for disease studies.

Catalog record

CREATORShihab HA, Rogers MF, Gough J, Mort M, Cooper DN, Day IN, Gaunt TR, Campbell C.
YEAR2015
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC4426838

Browse the full library · About the Tower of Babel

Full text

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

An integrative approach to predicting the functional effects of non-coding and coding sequence variation By Hashem A. Shihab, Mark F. Rogers, Julian Gough, Matthew Mort, David N. Cooper, Ian N. M. Day, Tom R. Gaunt, Colin Campbell Published in: Bioinformatics DOI: 10.1093/bioinformatics/btv009 ABSTRACT Motivation: Technological advances have enabled the identification of an increasingly large spectrum of single nucleotide variants within the human genome, many of which may be associated with monogenic disease or complex traits. Here, we propose an integrative approach, named FATHMM-MKL, to predict the functional consequences of both coding and non-coding sequence variants. Our method utilizes various genomic annotations, which have recently become available, and learns to weight the significance of each component annotation source. Results: We show that our method outperforms current state-of-the-art algorithms, CADD and GWAVA, when predicting the functional consequences of non-coding variants. In addition, FATHMM-MKL is comparable to the best of these algorithms when predicting the impact of coding variants. The method includes a confidence measure to rank order predictions. Availability and implementation: The FATHMM-MKL webserver is available at: http://fathmm.biocompute.org.uk Contact: [email protected] or [email protected] or [email protected] Supplementary information: Supplementary data are available at Bioinformatics online. 1 INTRODUCTION Rapid technological advances and the falling costs of next-generation sequencing technologies have accelerated the identification of single nucleotide variants (SNVs) in the human genome (The 1000 Genomes Project, 2010). Ascertaining which of these are functional, against a background of neutral SNVs, promises to improve our understanding of the molecular mechanisms underpinning disease and complex traits. There is a plethora of computational algorithms capable of predicting whether SNVs are deleterious (Thusberg et al., 2011). However, these algorithms are typically restricted to SNVs falling within protein-coding regions of the genome, with a particular focus on non-synonymous SNVs (nsSNVs). They are therefore incapable of assessing the consequences of a significant proportion of SNVs, given that the vast majority of SNV-trait associations fall within intergenic or intronic regions of the genome (Hindorff et al., 2009). In this article, we propose a method for predicting the functional impact of both coding and non-coding SNVs. Existing protein-based prediction algorithms tend to exploit evolutionary conservation when determining whether or not variants are functional. However, other potential sources for functional annotation are also now available from the Encyclopaedia of DNA Elements (ENCODE) consortium. Using several different technologies, ENCODE aims to identify all functional elements within the human genome, including transcribed non-coding RNAs, transcription factor binding sites and chromatin structure (The ENCODE Project Consortium, 2012). At the time of writing, ENCODE has produced approximately 1640 datasets describing the analysis of 24 different types of experiments in 147 cell lines, under various conditions (Qu and Fang, 2013). One of the major challenges faced by researchers is to successfully integrate the wealth of information contained within ENCODE to predict the functional consequences of SNVs. Following our previous work (Shihab et al., 2013b, a, 2014), we describe a machine learning approach (called FATHMM-MKL) that integrates functional annotations from ENCODE with nucleotide-based sequence conservation measures. When assessing the functional consequences of non-coding variants, we observe improved performance when compared with two recently proposed variant prediction algorithms: GWAVA (Ritchie et al., 2014) and CADD (Kircher et al., 2014), which as far as we are aware, are the only other proposed methods that can predict the functional consequences of non-coding variants. Furthermore, our method achieves comparable performance to CADD when predicting the functional impact of nsSNVs. A web-based implementation of FATHMM-MKL, including pre-computed predictions for the entire human genome and downloadable software, is available at http://fathmm.biocompute.org.uk. 2 METHODS AND MATERIALS 2.1 Datasets We assembled two distinct datasets: our pathogenic dataset was constructed using heritable germ-line mutations from the Human Gene Mutation Database (Stenson et al., 2014) (release 2013.4, subsequently denoted HGMD) and our control dataset was constructed using SNVs from the 1000 Genomes Project (The 1000 Genomes Project Consortium, 2012). This control dataset will contain many variants that are unannotated, and therefore it will likely contain some true positives. We will consider this issue in more detail in Section 3.3. Similarly, among positively labelled datapoints from the HGMD dataset, we could expect some true negatives: some neutral variants can be inherited alongside pathogenic variants due to genomic proximity, for example. Thus the positive class should be viewed as disease-associated variants that are enriched for functional impact as a set. The method we are to describe can be used with datasets for both inherited single nucleotide polymorphisms (SNPs) and somatic SNVs. The HGMD dataset is therefore an example of the former category. For our analysis, we restricted our control dataset to SNVs having a global minor allele frequency ≥1% and further removed those that were also present in our pathogenic dataset. In order to assess the predictive utility of our method in both coding and non-coding regions of the genome, we further split our pathogenic and control datasets according to whether or not the variant introduces an amino acid substitution. We used 10 feature groups, denoted [A–J], which could be predictive of disease association and are therefore used to annotate out datasets using a customized pipeline. These feature groups are more fully described in the Supplementary material, but a short description is as follows: 46-Way Sequence Conservation: based on multiple sequence alignment scores, at the nucleotide level, of 46 vertebrate genomes compared with the human genome.Histone Modifications (ChIP-Seq): based on ChIP-Seq peak calls for histone modifications.Transcription Factor Binding Sites (TFBS PeakSeq): based on PeakSeq peak calls for various transcription factors.Open Chromatin (DNase-Seq): based on DNase-Seq peak calls.100-Way Sequence Conservation: based on multiple sequence alignment scores, at the nucleotide level, of 100 vertebrate genomes compared with the human genome.GC Content: based on a single measure for GC content calculated using a span of five nucleotide bases from the UCSC Genome Browser.Open Chromatin (FAIRE): based on formaldehyde-assisted isolation of regulatory elements (FAIRE) peak calls.Transcription Factor Binding Sites (TFBS SPP): based on SPP peak calls for various transcription factors.Genome Segmentation: based on genome-segmentation states using a consensus merge of segmentations produced by the ChromHMM and Segway software.Footprints: based on annotations describing DNA footprints across cell types from ENCODE. 2.2 Data integration The resulting product of this data preparation is several large matrices comprising data from different feature groups, each of which can have different measurement scales. These feature groups can all indicate whether an SNV is functional or not, and hence we use a classifier based on multiple kernel learning (MKL). In MKL, different types of input data are encoded into kernel matrices, which quantify the similarity of data objects. A number of different methods have been proposed for deriving kernel matrices for different types of data objects, including data with discrete and continuous values, sequence data and graph data (Shawe-Taylor and Cristianini, 2004). With MKL, each constituent type of data is encoded into a corresponding base kernel Kℓ (where ℓ=1,…,p if there are p feature groups), from which we can derive a composite kernel matrix K=∑ℓ=1pλℓKℓ where ∑ℓ=1pλℓ=1 and λℓ≥0. The λℓ are kernel weights. This composite kernel can then be used with a kernel-based classifier, such as a support vector machine (SVM) (Campbell and Ying, 2011), which was the classifier used here. During the training phase, this approach requires determination of the learning parameters

Read the complete text