Improved identification of concordant and discordant gene expression signatures using an updated rank-rank hypergeometric overlap approach.
Cahill KM, Huo Z, Tseng GC, Logan RW, Seney ML. · 2018
Read Improved identification of concordant and discordant gene expression signatures using an updated rank-rank hypergeometric overlap approach. by Cahill KM, Huo Z, Tseng GC, Logan RW, Seney ML. (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-pmc6018631.txt.
Introduction
Cahill and co-authors present an updated method for gene studies. Their article appeared in Scientific Reports in 2018. The method compares ranked lists of gene activity. It looks for genes that change in the same direction. It also finds genes that change in opposite directions.
Most tools rely on strict significance cutoffs. Those cutoffs can miss real but subtle shared patterns. The updated approach, called RRHO2, avoids a single cutoff. It scans overlap across a full gradient of ranks. Tests used simulated data and real brain data. Data came from postmortem brains in psychiatric disorders. The package also offers clearer plots. Readers get a practical guide to finding shared expression signatures.
Catalog record
| CREATOR | Cahill KM, Huo Z, Tseng GC, Logan RW, Seney ML. |
|---|---|
| YEAR | 2018 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63 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:
Improved identification of concordant and discordant gene expression signatures using an updated rank-rank hypergeometric overlap approach By Kelly M. Cahill, Zhiguang Huo, George C. Tseng, Ryan W. Logan, Marianne L. Seney Published in: Scientific Reports DOI: 10.1038/s41598-018-27903-2 ABSTRACT Recent advances in large-scale gene expression profiling necessitate concurrent development of biostatistical approaches to reveal meaningful biological relationships. Most analyses rely on significance thresholds for identifying differentially expressed genes. We use an approach to compare gene expression datasets using ‘threshold-free’ comparisons. Significance cut-offs to identify genes shared between datasets may be too stringent and may miss concordant patterns of gene expression with potential biological relevance. A threshold-free approach gaining popularity in several research areas, including neuroscience, is Rank–Rank Hypergeometric Overlap (RRHO). Genes are ranked by their p-value and effect size direction, and ranked lists are compared to identify significantly overlapping genes across a continuous significance gradient rather than at a single arbitrary cut-off. We have updated the previous RRHO analysis by accurately detecting overlap of genes changed in the same and opposite directions between two datasets. Here, we use simulated and real data to show the drawbacks of the previous algorithm as well as the utility of our new algorithm. For example, we show the power of detecting discordant transcriptional patterns in the postmortem brain of subjects with psychiatric disorders. The new R package, RRHO2, offers a new, more intuitive visualization of concordant and discordant gene overlap. INTRODUCTION Comparing patterns of gene expression between experimental groups is often useful for exploring common molecular pathways potentially involved in specific biological processes. Current approaches typically use correlations or enrichment between multiple sets of genes employing statistical cutoffs (e.g., p < 0.05 or 5% false discovery rate) to derive a list of differentially expressed genes. In large-scale gene expression experiments (microarrays, RNA-seq, etc), gene sets are narrowed from thousands to hundreds or maybe tens of genes, which represent the most significant genes with presumably the highest relevance. These gene sets are then characterized based on pathways, functions, or structural characteristics from popular annotated databases. For enrichment, statistical significance is typically determined by the hyper-geometric distribution, or one-sided Fisher’s exact test. While these are intuitive and readily implemented, genome-wide sequencing experiments generate a continuous range of expression differences of thousands of genes, such that ‘hard’ arbitrary cutoffs could impact the veracity of significance (e.g., conservative vs. lenient), consequently discarding subtle changes and relevant biological information. Often, researchers aim to compare two separate differential gene expression datasets to identify overlap trends. For instance, one might examine differential gene expression between stressed and non-stressed mice in two different brain regions. In this example, there are two separate lists of differentially expressed genes (one for each brain region). To identify genes that are differentially expressed in both brain regions, genes are identified using a strict significance cutoff in each dataset (e.g., p < 0.05 in both); lists of genes are then compared to reveal overlapping pathways between groups. While these approaches have proved valuable, they are greatly impacted by experimental design, sample size, and perturbation, and may miss relevant biological information. A complementary approach to using significance cutoffs when comparing two differential expression datasets is a threshold free algorithm called rank-rank hyper-geometric overlap (RRHO). RRHO enables the direct comparison of separate gene sets to reveal overlapping trends. For each dataset, RRHO ranks the entire gene list by differential expression p-values and effect size direction. For each dataset, these ranks are arranged vertically or horizontally to graphically display the significance of the overlap intensity. Thus, the algorithm steps through each ranked gene to obtain significance values representing the number of overlapping genes and the resulting significance matrix is visualized using a heatmap. Each point represents the significance of the number of overlapping genes via the hyper-geometric distribution1. The final heatmap of the RRHO represents potentially overlapping biologically relevant signatures. RRHO can be used to compare differential gene expression profiles between multiple species, different microarray platforms, model systems, and experimental perturbations, such as genetic manipulations or drug treatments. RRHO provides an intuitive and comprehensive tool for identifying common patterns of gene expression between multiple datasets combined with graphical representation which summarizes the intensity of these patterns. The software for conducting RRHO is publically available as a Bioconductor R package and other web-based applications. Applications of RRHO have been widely used to determine gene enrichment between multiple datasets where the intention is to reveal common molecular signatures of independent groups2–6. The input for RRHO requires gene significance values derived from group comparisons and including the effect size direction is valuable for the assessment and interpretation of the overlaps. Currently, there are two representations for RRHO: Enrichment and Two-sided. For Enrichment, each pixel in the heatmap represents higher than expected overlap between gene sets and the overlaps are represented in quadrants, where the bottom left and top right represent up-regulated and down-regulated genes in common between both gene sets, respectively (i.e., concordance). This is the option most frequently applied, and most studies implementing this approach are designed to discover concordance. The representation of the top left and bottom right quadrants was intended to identify overlapping genes with opposite expression patterns (i.e., discordance). The algorithm originally proposed by Plaiser and colleagues1 suggests ‘hotspots’ occurring within these other quadrants adequately denotes genes up-regulated in the one dataset and down-regulated in the other. However, the interpretation of these quadrants has been more challenging. Similarly, the Two-sided approach leads to ambiguous results due to the potential inverse of positive and negative heatmap values. Values in the top left and bottom right quadrants must be interpreted with caution. We propose a modified algorithm to improve the efficiency, representation, and interpretation of these powerful RRHO approaches. First, we demonstrate the limitations of the current RRHO approaches to reveal discordant gene signatures. Second, we update the RRHO algorithms to enhance the Enrichment and Two-sided approaches in order to provide more intuitive representations, which should lead to improved implementation of these applications. Third, we propose an additional approach to the RRHO package which uses the Fisher exact test effect size to rank the genes rather than p-values. We compared the performance of our Stratified representation with the Enrichment and Two-sided algorithms using simulated and experimental datasets. Lastly, we made the RRHO2 package publicly available on Bioconductor. METHODS Original RRHO implementation In order to compare the concurrence of the top differential expression genes from two independent studies, S1 and S2, the RRHO algorithm utilizes the hypergeometric distribution to determine the degree of overlap between particular subsets of genes. The hypergeometric distribution is as follows:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h(k;s,M,N)=\,\frac{(\begin{array}{c}M\\ k\end{array})(\begin{array}{c}N-M\\ s-k\end{array})}{(\begin{array}{c}N\\ s\end{array})}$$\end{document}h(k;s,M,N)=(Mk)(N−Ms−k)(Ns) The distribution describes the number of successes in a sequence of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s$$\end{document}s draws from a finite population \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N$$\end{document}N without replacement. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k$$\end{document}k is the number of successes in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s$$\end{document}s draws and M is the number of success in the whole population. The expected number of successes from the hypergeometric distribution is:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E(k)=\,\bar{k}=s\frac{M}{N}$$\end{document}E(k)=k¯=sMN For