receptLoss
Daniel G. Piqué, John Greally, and Jessica Mar
May 19, 2021
receptLoss is an R package designed to identify novel nuclear hormone receptors (NHRs) whose expression levels in cancers could serve as biomarkers for patient survival.
By utilizing both expression data from both tumor and normal tissue, receptLoss provides biological context to the process of tumor subclassification that is lacking in existing methods that rely solely on expression patterns in tumor tissue.
receptLoss is complementary to oncomix. Whereas oncomix detects genes that gain expression in subsets of tumors relative to normal tissue, receptLoss detects genes that lose expression in subsets of tumors relative to normal tissue.
Installation
## Install the development version from GitHub
devtools::install_github("dpique/receptLoss",
build_opts=c("--no-resave-data", "--no-manual"),
build_vignettes=TRUE)Usage
receptLoss consists of 2 main functions:
receptLoss()takes in 2 matrices of gene expression data, one from tumor and one from adjacent normal tissue. The output is a matrix with rows representing genes and columns representing summary statistics.plotReceptLoss()generates a histogram visualization of the distribution of the gene desired by the user.
We begin by simulating two gene expression data matrices, one from tumor and the other from normal tissue.
library(receptLoss)
library(dplyr)
library(ggplot2)
set.seed(100)
## Simulate matrix of expression values from
## 10 genes measured in both normal tissue and
## tumor tissue in 100 patients
exprMatrNml <- matrix(abs(rnorm(100, mean=2.5)), nrow=10)
exprMatrTum <- matrix(abs(rnorm(1000)), nrow=10)
geneNames <- paste0(letters[seq_len(nrow(exprMatrNml))],
seq_len(nrow(exprMatrNml)))
rownames(exprMatrNml) <- rownames(exprMatrTum) <- geneNamesexprMatrNml and exprMatrTum are \(m \times n\) matrices containing gene expression data from normal and tumor tissue, respectively, with \(m\) genes as rows and \(n\) patients as columns. The row names of these matrices are the gene names.
These two matrices should have the same number of rows (ie genes), with genes listed in the same order between the two matrices. However, they don’t have to have the same number of columns (ie patients).
To run receptLoss(), we also define 2 parameters:
nSdBelowis an integer value that places a lower boundary (i.e.lowerBound, shown as the pink ‘B’ in image below) \(n\) standard deviations below the mean of each gene’s expression levels in normal tissue (dotted pink curve below). The largernSdBelowis, the smaller (i.e. further to the left) thelowerBoundbecomes.- We recommend setting
nSdBelow=2, as ~97.7% of the normal tissue expression data should be greater than thelowerBound(assuming the expression data from normal tissue is distributed as a Gaussian).
- We recommend setting
minPropPerGroup- a numeric value between \((0,0.5)\) indicating the minimum proportion of tumor samples desired within each of the two tumor subgroups defined by thelowerBound. Determines the value ofmeetsMinPropPerGrp(either TRUE or FALSE) in the output.- We recommend setting
minPropPerGroup=0.20. Values close to 0 may result in the inclusion of genes that subdivide tumors into very unequally-sized subgroups. Values closer to 0.5 will identify genes that subdivide tumors into nearly equal-sized groups and may be unnecessarily restrictive.
- We recommend setting
<img src=“fig_2_1.png” alt=“fig 2”, width=“60%”>
nSdBelow <- 2
minPropPerGroup <- .2
rl <- receptLoss(exprMatrNml, exprMatrTum, nSdBelow, minPropPerGroup)
head(rl)
#> # A tibble: 6 x 7
#> geneNm lowerBound propTumLessThBound muAb muBl deltaMu meetsMinPropPerGrp
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <lgl>
#> 1 f6 0.928 0.58 1.52 0.425 1.10 TRUE
#> 2 b2 0.538 0.38 1.30 0.258 1.05 TRUE
#> 3 i9 0.379 0.24 1.10 0.203 0.893 TRUE
#> 4 g7 0.805 0.57 1.30 0.405 0.891 TRUE
#> 5 d4 0.359 0.32 1.04 0.174 0.866 TRUE
#> 6 c3 0.554 0.41 1.12 0.290 0.826 TRUEThe output of receptLoss() is an \(m\times7\) matrix, with \(m\) equaling the number of genes. The 7 columns are as follows:
geneNm- the gene namelowerBound(\(B\)) - the valuenSdBelowthe mean of the normal tissue expression data. Can be expressed as \[B=\mu_N - \sigma_N * n_{sdBelow},\] where \(\mu_N\) is the mean of the normal tissue expression data, \(\sigma_N\) is the standard deviation of the normal tissue expression data, and \(n_{sdBelow}\) is the valuenSdBelowset by the user.propTumLessThBound(\(\pi_L\)) - the proportion of tumor samples with expression levels less thanlowerBound. Can be expressed as: \[\pi_L =\frac{1}{N_T}\sum_{j=1}^{N_T} \Bigg\{ \begin{array}{ll} 1,~ if~x_{j} < lowerBound \\ 0, ~ otherwise \end{array}, \] where \(x_{j}\) is the \(j^{th}\) tumor sample and \(N_T\) is the total number of tumor samples.muAb(\(\mu_A\)) - “mu above”, the arithmetic mean across expression values from tumors greater than (ie above) thelowerBound.muBl(\(\mu_B\)) - “mu below”, the arithmetic mean across expression values from tumors less than (ie below) thelowerBound.deltaMu(\(\Delta\mu\)) - equal to \(\mu_A - \mu_B\). The rows in the output matrix are sorted in descending order by thedeltaMustatistic, which indicates the degree of separation between the two tumor subgroups. HigherdeltaMuvalues indicate tumor subgroups that are more cleanly separated and more likely to constitute a bimodal distribution within the tumor samples.meetsMinPropPerGrp- a logical indicating whether the proportion of samples in each group is greater than that set byminPropPerGroup. If \(min(\pi_L, 1-\pi_L) >\)minPropPerGroup, thenmeetsMinPropPerGrpisTRUE; otherwise, it isFALSE. Genes for whichmeetsMinPropPerGrpequalsFALSEcan be filtered out - they do not have a sufficient proportion of tumors in each group to permit useful tumor subgrouping.
Visualization
Let’s take the top-ranked gene and plot its distribution.
clrs <- c("#E78AC3", "#8DA0CB")
tryCatch({plotReceptLoss(exprMatrNml, exprMatrTum, rl,
geneName=as.character(rl[1,1]), clrs=clrs)},
warning=function(cond){
knitr::include_graphics("rl_fig.png")
}, error=function(cond){
knitr::include_graphics("rl_fig.png")
}
)
Here’s what this graph is showing us:
The x-axis represents RNA expression values, with lower values toward the left and larger values (i.e. higher expression) toward the right. The y-axis represents density. The name of the gene (“f6”) is shown in the upper left of the plot.
The dotted curve represents a Gaussian distribution fit to the expression data from normal tissue, and the blue histogram represents expression data from tumor tissue.
The pink vertical line corresponds to the
lowerBoundfor the expression data from normal tissue.Since most normal tissue expresses the RNA above the
lowerBound, any tumors that express the RNA below this value have lost RNA expression relative to normal tissue. Thus, thelowerBoundforms a boundary between 2 tumor subgroups that either have or have not lost RNA expression relative to normal tissue.
Nuclear Hormone Receptor (NHR) filtering
The question that inspired this package was whether the loss of expression of any of the ~50 NHRs (beyond the well-known estrogen, progesterone, and androgen NHRs) in uterine tumors was associated with differences in patient survival. NHRs might not only serve as survival biomarkers but also as drug targets, as their activity can be modulated by small molecules that resemble their hormonal ligands.
To facilitate the application of this question to additional cancer types, a list of all NHRs is included in this package as the object nhrs.
This object facilitates filtering of NHRs from a matrix of gene expression data, as it contains several commonly-used gene identifiers (e.g. HGNC symbol, HGNC ID, Entrez ID, and Ensembl ID) for the NHRs that might be found in different RNA expression datasets.
The source code for generating nhrs is available in “data-raw/nhrs.R”.
receptLoss::nhrs
#> # A tibble: 54 x 6
#> hgnc_symbol hgnc_id hgnc_name entrez_gene_id ensembl_gene_id synonyms
#> <chr> <dbl> <chr> <dbl> <chr> <chr>
#> 1 NR0B1 7960 nuclear rec… 190 ENSG00000169297 AHC|DSS|AHCH…
#> 2 NR0B2 7961 nuclear rec… 8431 ENSG00000131910 Small hetero…
#> 3 THRA 11796 thyroid hor… 7067 ENSG00000126351 AR7|EAR-7.1/…
#> 4 THRB 11799 thyroid hor… 7068 ENSG00000151090 THR1|THRB1|T…
#> 5 RARA 9864 retinoic ac… 5914 ENSG00000131759 RAR alpha 1|…
#> 6 RARB 9865 retinoic ac… 5915 ENSG00000077092 HAP|HBV-acti…
#> 7 RARG 9866 retinoic ac… 5916 ENSG00000172819 RARC|RAR&gam…
#> 8 PPARA 9232 peroxisome … 5465 ENSG00000186951 NUC1|nuclear…
#> 9 PPARD 9235 peroxisome … 5467 ENSG00000112033 NUCII|PPAR&b…
#> 10 PPARG 9236 peroxisome … 5468 ENSG00000132170 PPARG1|PPARG…
#> # … with 44 more rowsConclusions
receptLossidentifies genes that subclassify tumors based on their RNA expression levels relative to normal tissue. The genes are ranked by their \(\Delta\mu\) statistic which reflects a measure of the cleaness of separation (ie bimodality) between the two tumor subgroups.receptLosscan be expanded for use with a variety of tumors, genes (e.g. to identify novel candidate tumor suppressors), biological data (e.g. miRNA, protein expression), and even non-biological data types where you have numeric data from two groups (one normal group and one abnormal group) and where subgroup identification is desired within the abnormal groupreceptLossis particularly useful when there are a large number of tumor samples (hundreds) relative to normal samples (dozens), as is the case in several cancer databases, including the uterine cancer database from the Cancer Genome Atlas/Genomic Data Commons. By assuming that the normal expression data are distributed as a single Gaussian,receptLosscan subclassify large numbers of tumors even in the presence of small numbers of normal tissue samples.
Please contact me at daniel.pique@med.einstein.yu.edu with any suggestions, questions, or comments. Thank you!
Display this vignette
vignette("receptLoss") Session Info
sessionInfo()
#> R version 4.1.0 (2021-05-18)
#> Platform: x86_64-pc-linux-gnu (64-bit)
#> Running under: Ubuntu 20.04.2 LTS
#>
#> Matrix products: default
#> BLAS: /home/biocbuild/bbs-3.13-bioc/R/lib/libRblas.so
#> LAPACK: /home/biocbuild/bbs-3.13-bioc/R/lib/libRlapack.so
#>
#> locale:
#> [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
#> [3] LC_TIME=en_GB LC_COLLATE=C
#> [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
#> [7] LC_PAPER=en_US.UTF-8 LC_NAME=C
#> [9] LC_ADDRESS=C LC_TELEPHONE=C
#> [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
#>
#> attached base packages:
#> [1] stats graphics grDevices utils datasets methods base
#>
#> other attached packages:
#> [1] ggplot2_3.3.3 dplyr_1.0.6 receptLoss_1.4.0
#>
#> loaded via a namespace (and not attached):
#> [1] SummarizedExperiment_1.22.0 tidyselect_1.1.1
#> [3] xfun_0.23 bslib_0.2.5.1
#> [5] purrr_0.3.4 lattice_0.20-44
#> [7] colorspace_2.0-1 vctrs_0.3.8
#> [9] generics_0.1.0 htmltools_0.5.1.1
#> [11] stats4_4.1.0 yaml_2.2.1
#> [13] utf8_1.2.1 rlang_0.4.11
#> [15] jquerylib_0.1.4 pillar_1.6.1
#> [17] withr_2.4.2 glue_1.4.2
#> [19] DBI_1.1.1 BiocGenerics_0.38.0
#> [21] matrixStats_0.58.0 GenomeInfoDbData_1.2.6
#> [23] lifecycle_1.0.0 stringr_1.4.0
#> [25] MatrixGenerics_1.4.0 zlibbioc_1.38.0
#> [27] munsell_0.5.0 gtable_0.3.0
#> [29] evaluate_0.14 labeling_0.4.2
#> [31] Biobase_2.52.0 knitr_1.33
#> [33] IRanges_2.26.0 ps_1.6.0
#> [35] GenomeInfoDb_1.28.0 parallel_4.1.0
#> [37] fansi_0.4.2 highr_0.9
#> [39] scales_1.1.1 DelayedArray_0.18.0
#> [41] S4Vectors_0.30.0 jsonlite_1.7.2
#> [43] XVector_0.32.0 png_0.1-7
#> [45] digest_0.6.27 stringi_1.6.2
#> [47] GenomicRanges_1.44.0 grid_4.1.0
#> [49] cli_2.5.0 tools_4.1.0
#> [51] bitops_1.0-7 magrittr_2.0.1
#> [53] sass_0.4.0 RCurl_1.98-1.3
#> [55] tibble_3.1.2 crayon_1.4.1
#> [57] tidyr_1.1.3 pkgconfig_2.0.3
#> [59] ellipsis_0.3.2 Matrix_1.3-3
#> [61] rstudioapi_0.13 assertthat_0.2.1
#> [63] rmarkdown_2.8 R6_2.5.0
#> [65] compiler_4.1.0