Single sample scoring of molecular phenotypes.
Foroutan M, Bhuva DD, Lyu R, Horan K, Cursons J, Davis MJ. · 2018
Read Single sample scoring of molecular phenotypes. by Foroutan M, Bhuva DD, Lyu R, Horan K, Cursons J, Davis MJ. (2018) 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-pmc6219008.txt.
Introduction
Foroutan and co-authors present a method to score single samples by gene sets. Gene set scores show how well a sample matches a chosen signature. Most methods score a sample using all other samples. That makes scores unstable in small datasets. The work appeared in BMC Bioinformatics.
Their method, called singscore, needs no background samples. Scores stay stable no matter how many samples are scored. The team tested it on many cancer datasets. They compared it with GSVA, z-score, PLAGE, and ssGSEA. Rival scores grew unstable when samples were few. Singscore matched the best rivals on power and accuracy. It also ships as an R package with visual tools.
Catalog record
| CREATOR | Foroutan M, Bhuva DD, Lyu R, Horan K, Cursons J, Davis MJ. |
|---|---|
| YEAR | 2018 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 46 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Single sample scoring of molecular phenotypes By Momeneh Foroutan, Dharmesh D. Bhuva, Ruqian Lyu, Kristy Horan, Joseph Cursons, Melissa J. Davis Published in: BMC Bioinformatics DOI: 10.1186/s12859-018-2435-4 ABSTRACT Background Gene set scoring provides a useful approach for quantifying concordance between sample transcriptomes and selected molecular signatures. Most methods use information from all samples to score an individual sample, leading to unstable scores in small data sets and introducing biases from sample composition (e.g. varying numbers of samples for different cancer subtypes). To address these issues, we have developed a truly single sample scoring method, and associated R/Bioconductor package singscore (https://bioconductor.org/packages/singscore). Results We use multiple cancer data sets to compare singscore against widely-used methods, including GSVA, z-score, PLAGE, and ssGSEA. Our approach does not depend upon background samples and scores are thus stable regardless of the composition and number of samples being scored. In contrast, scores obtained by GSVA, z-score, PLAGE and ssGSEA can be unstable when less data are available (NS < 25). The singscore method performs as well as the best performing methods in terms of power, recall, false positive rate and computational time, and provides consistently high and balanced performance across all these criteria. To enhance the impact and utility of our method, we have also included a set of functions implementing visual analysis and diagnostics to support the exploration of molecular phenotypes in single samples and across populations of data. Conclusions The singscore method described here functions independent of sample composition in gene expression data and thus it provides stable scores, which are particularly useful for small data sets or data integration. Singscore performs well across all performance criteria, and includes a suite of powerful visualization functions to assist in the interpretation of results. This method performs as well as or better than other scoring approaches in terms of its power to distinguish samples with distinct biology and its ability to call true differential gene sets between two conditions. These scores can be used for dimensional reduction of transcriptomic data and the phenotypic landscapes obtained by scoring samples against multiple molecular signatures may provide insights for sample stratification. Electronic supplementary material The online version of this article (10.1186/s12859-018-2435-4) contains supplementary material, which is available to authorized users. BACKGROUND Several approaches have been developed to score individual samples against molecular signatures (or gene sets), including: ssGSEA (single sample gene set enrichment analysis) [1], GSVA (gene set variation analysis) [2], PLAGE (pathway level analysis of gene expression) [3] and combining z-scores [4]. Hänzelmann et al. (2013) implemented all four of these methods within the R/Bioconductor package GSVA and performed a detailed comparison [2, 5]. It should be noted that GSVA, PLAGE and z-scores use data from all samples in the very first step to estimate gene distributions; GSVA performs kernel density estimation of the expression profile for each gene across all samples, while PLAGE and z-scores perform standardisation. Although the ssGSEA implementation in the GSVA package normalises the scores across samples, this is the final step and it can be disabled. The ssGSEA method is also implemented through the GenePattern web-tool [6] which does not normalise scores by default. Some methods make assumptions about the data which may be unsuitable in certain cases, for instance, PLAGE and combined z-scores are parametric methods that assume normality of expression profiles, while the combined z-scores method additionally makes an independence assumption for genes in a gene set [7]. Here, we introduce a rank-based single sample scoring method, singscore. Using breast cancer data and several gene expression signatures we compare our approach to the methods listed above. The singscore method is simple, making the scores directly interpretable (as a normalised mean percentile rank), and our comparisons show that it is not only fast, but it also produces stable and reproducible scores regardless of the composition and number of samples within the data. Finally, we include examples from breast cancer gene expression data to show visualisation options and demonstrate the application of the singscore method for molecular phenotyping in a clinical context. METHODS The singscore method For a sample transcriptome which has been corrected for technical within-sample bias (i.e. RPKM, TPM, or RSEM data for RNA-seq after filtering for genes with low-counts), genes are ranked by increasing mRNA abundance. For bidirectional gene signatures (with separate up- and down- regulated gene sets) or unidirectional signatures with known direction (e.g. all genes are up-regulated), genes are ranked based on their transcript abundance in increasing order for the up-set and in decreasing order for the down-set. Mean ranks are separately normalised relative to the theoretical minimum and maximum values, centered on zero and then summed to provide the score (i.e. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\overline{S}}_{total,i}={\overline{S}}_{up,i}+{\overline{S}}_{down,i} $$\end{document}S¯total,i=S¯up,i+S¯down,i) which ranges between − 1 and 1. A sample with a high score can be interpreted as having a transcriptome which is concordant to the specified signature, and scores reflect the relative mean percentile rank of the target gene sets within each sample. The score (S) and normalised score (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \overline{S} $$\end{document}S¯) are defined as:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {S}_{dir,i}=\left(\frac{\sum_g{R}_{dir,i}^g}{N_{dir,i}}\right) $$\end{document}Sdir,i=∑gRdir,igNdir,i2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\overline{S}}_{dir,i}=\frac{\left({S}_{dir,i}-{S}_{\mathit{\min},i}\right)}{S_{\mathit{\max},i}-{S}_{\mathit{\min},i}} $$\end{document}S¯dir,i=Sdir,i−Smin,iSmax,i−Smin,i Where:dir is the gene set direction (i.e. expected up- or down- regulated genes);Sdir, i is the score for sample i against the directed gene set;\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {R}_{dir,i}^g $$\end{document}Rdir,ig is the rank of gene g in the directed gene set (increasing transcript abundance for expected up-regulated genes and decreasing abundance for expected down-regulated genes);Ndir, i is the number of genes in the expected up- or down-regulated gene set that are observed within the data (i.e. signature genes not present within the RNA abundance data are excluded);\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\overline{S}}_{dir,i} $$\end{document}S¯dir,i is the normalised score for sample i against genes in the signature, and;Smin, i and Smax, i are the theoretical minimum and maximum mean ranks which can be derived from an arithmetic sum (assuming unique ranks); for a series of n numbers starting at a1 and with a constant difference d, the sum can be calculated as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \left(\raisebox{1ex}{$n$}\!\left/ \!\raisebox{-1ex}{$2$}\right.\right)\left(2{a}_1+\left(n-1\right)d\right) $$\end{document}n22a1+n−1d. Setting a1 = 1, d = 1, n = Ndir, i and dividing through by Ndir, i to obtain the mean: 3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {S}_{\mathit{\min},i}=\frac{\left({N}_{dir,i}+1\right)}{2} $$\end{document}Smin,i=Ndir,i+12 Similarly, the maximum value can be obtained by setting a1 = (Ntotal – Ndir):4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {S}_{\mathit{\max},i}=\frac{\left(2{N}_{total,i}-{N}_{dir,i}+1\right)}{2} $$\end{document}Smax,i=2Ntotal,i−Ndir,i+12 Where:Ntotal, i is the total number of genes in sample i. For undirected gene signatures (without a known or expected direction for changes in expression), we compute the average, absolute, median-centred rank (Eqs. 5 & 6). As the direction of genes in the gene-set are unknown, this method only provides evidence of enrichment of more-extreme gene ranks in either direction, not necessarily information about the specific direction (Additional file 1: Figure S1). Resulting scores range between 0 and 1; as shown in Eqs. 7 & 8, the normalisation step is altered:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs}