Reproducible Research in R / Bioconductor
Martin Morgan
October 31, 2016
Introduction
This workshop uses data from a gene-level RNA-seq experiment involving airway smooth muscle cells; details are provided in the vignette accompanying the airway package. The original data is from Himes et al., “RNA-Seq Transcriptome Profiling Identifies CRISPLD2 as a Glucocorticoid Responsive Gene that Modulates Cytokine Function in Airway Smooth Muscle Cells.” PLoS One. 2014 Jun 13;9(6):e99625. PMID: 24926665 GEO: GSE52778. From the Abstract: “Using RNA-Seq […] we characterized transcriptomic changes in four primary human ASM cell lines that were treated with dexamethasone - a potent synthetic glucocorticoid (1 micromolar for 18 hours).”
We join the analysis after the sequencing, read aligment, and summary of aligned reads to a table of counts of reads overlapping regions of interest (genes) in each sample. We focus on a subset of the experiment, with 4 cell lines each treated with dexamethasone or a control.
Setup (not necessary during the workshop)
We’ll use two packages from Bioconductor. These depend in turn on several other packages. The packages and their dependencies are already installed on the Amazon Machine Instance (AMI) used in the course. For your own computers after the course, install packages with
source("https://bioconductor.org/biocLite.R")
biocLite(c("DESeq2", "org.Hs.eg.db"))Installation needs to be performed once per computer, not every time the packages are used.
We use two data files in the analysis. The data files are in the Sydney2016 github repository. Install the github repository with
biocLite("Bioconductor/Syndey2016")Once the package is installed, the location of the files (to be used in file.choose(), below) is given by
system.file(package="Sydney2016", "extdata")Data input
The first challenge is to input the data. We start with the ‘phenotypic’ data, describing the samples used in the experiment. The data is a simple table of 8 rows and several columns; it could be created in Excel and exported as a tab-delimited file. Find the location of the file on your Amazon machine instance
colDataFile <- file.chooose() # find 'airway-colData.tab'and read the data in to R using the read.table() function. The data is small enough to be viewed in the R session by typing the name of the variable) or in RStudio (by using View() or double-clicking on the variable in the ‘Environment’ tab).
colData <- read.table(colDataFile)
colData## SampleName cell dex albut Run avgLength Experiment
## SRR1039508 GSM1275862 N61311 untrt untrt SRR1039508 126 SRX384345
## SRR1039509 GSM1275863 N61311 trt untrt SRR1039509 126 SRX384346
## SRR1039512 GSM1275866 N052611 untrt untrt SRR1039512 126 SRX384349
## SRR1039513 GSM1275867 N052611 trt untrt SRR1039513 87 SRX384350
## SRR1039516 GSM1275870 N080611 untrt untrt SRR1039516 120 SRX384353
## SRR1039517 GSM1275871 N080611 trt untrt SRR1039517 126 SRX384354
## SRR1039520 GSM1275874 N061011 untrt untrt SRR1039520 101 SRX384357
## SRR1039521 GSM1275875 N061011 trt untrt SRR1039521 98 SRX384358
## Sample BioSample
## SRR1039508 SRS508568 SAMN02422669
## SRR1039509 SRS508567 SAMN02422675
## SRR1039512 SRS508571 SAMN02422678
## SRR1039513 SRS508572 SAMN02422670
## SRR1039516 SRS508575 SAMN02422682
## SRR1039517 SRS508576 SAMN02422673
## SRR1039520 SRS508579 SAMN02422683
## SRR1039521 SRS508580 SAMN02422677This should go smoothly; with real data one often needs to spend considerable time adjusting arguments to read.table() to account for the presence of a header, row names, comments, etc.
The next challenge is to input the expression estimates. This is a matrix of rows representing regions of interest (genes) and columns representing samples. Entries in the matrix are the number of reads overlapping each region in each sample. It is important that the values are raw counts, rather than scaled measures such as FPKM. Find the file
assayFile <- file.chooose() # find 'airway-assay.tab'Input the data and use head() to view the first few rows of the data.
assay <- read.table(assayFile)
head(assay)## SRR1039508 SRR1039509 SRR1039512 SRR1039513 SRR1039516
## ENSG00000000003 679 448 873 408 1138
## ENSG00000000419 467 515 621 365 587
## ENSG00000000457 260 211 263 164 245
## ENSG00000000460 60 55 40 35 78
## ENSG00000000938 0 0 2 0 1
## ENSG00000000971 3251 3679 6177 4252 6721
## SRR1039517 SRR1039520 SRR1039521
## ENSG00000000003 1047 770 572
## ENSG00000000419 799 417 508
## ENSG00000000457 331 233 229
## ENSG00000000460 63 76 60
## ENSG00000000938 0 0 0
## ENSG00000000971 11027 5176 7995Exploration
Calculate the ‘library size’ (total number of mapped reads) of each sample using colSums()
colSums(assay)## SRR1039508 SRR1039509 SRR1039512 SRR1039513 SRR1039516 SRR1039517
## 20637971 18809481 25348649 15163415 24448408 30818215
## SRR1039520 SRR1039521
## 19126151 21164133Create a density plot of the average asinh-transformed (asinh is log-like, expect near zero) read counts of each gene using the following series of commands.
plot(density(rowMeans(asinh(assay))))
Multi-dimensional scaling (MDS) is a dimensionality reduction method that takes vectors in n-space and projects them into two (or more) dimensions. Use the dist() function to calculate the (Euclidean) distance bewteen each sample, and the base R function cmdscale() to perform MDS on the distance matrix. We can use plot() to visualize the results and see the approximate location of each of the 8 samples. Use the argument col to color the points based on cell line (colData$cell) or experimental treatment colData$dex.
d <- dist(t(asinh(assay)))
plot(cmdscale(d), pch=19, cex=2)
plot(cmdscale(d), pch=19, cex=2, col=colData$cell)
plot(cmdscale(d), pch=19, cex=2, col=colData$dex)
Note that cell lines are relatively similar to one another. This suggests that cell line should be used as a covariate in subsequent analysis.
Differential expression
We will use the DESeq2 package for differential expression analysis; other choices are possible, notably edgeR and limma.
The analysis starts by providing the expression count data, a description of the experiment, and a ‘model’ that describes the statistical relationship we’d like to estimate. For our model and based in part on the exploratory analysis of the previous section, we suppose that count is determined cell line and dexamethasone treatment. We include cell line primarily as a covariate; our primary interest is in dexamethasone.
library(DESeq2)
dds <- DESeqDataSetFromMatrix(assay, colData, ~ cell + dex)The analysis is extremely straight-forward to invoke, but the calculations involve a number of sophisticated statistical issues, including:
- Robust estimate of library size
- Use of a negative binomial model to describe the relationship between experimental design and counts
- Use of moderated test statistics that balance gene-level parameter estimates with experiment wide estimates
- Data-driven approaches to outlier detection
- Filtering and control for multiple testing
The code is invoked as:
dds <- DESeq(dds)## estimating size factors## estimating dispersions## gene-wise dispersion estimates## mean-dispersion relationship## final dispersion estimates## fitting model and testingdds## class: DESeqDataSet
## dim: 33469 8
## metadata(1): version
## assays(3): counts mu cooks
## rownames(33469): ENSG00000000003 ENSG00000000419 ...
## ENSG00000273492 ENSG00000273493
## rowData names(46): baseMean baseVar ... deviance maxCooks
## colnames(8): SRR1039508 SRR1039509 ... SRR1039520 SRR1039521
## colData names(10): SampleName cell ... BioSample sizeFactorThe DESeq() function returns an object that can be used as a starting point for further analysis, for instance generating a ‘top table’ of differentially expressed genes, orderd by adjusted (for multiple comparison) P values.
result <- results(dds)
result## log2 fold change (MAP): dex untrt vs trt
## Wald test p-value: dex untrt vs trt
## DataFrame with 33469 rows and 6 columns
## baseMean log2FoldChange lfcSE stat
## <numeric> <numeric> <numeric> <numeric>
## ENSG00000000003 708.6021697 0.37415246 0.09884435 3.7852692
## ENSG00000000419 520.2979006 -0.20206175 0.10974241 -1.8412367
## ENSG00000000457 237.1630368 -0.03616686 0.13834540 -0.2614244
## ENSG00000000460 57.9326331 0.08445399 0.24990709 0.3379415
## ENSG00000000938 0.3180984 0.08413901 0.15133424 0.5559813
## ... ... ... ... ...
## ENSG00000273487 8.1632350 -0.55007238 0.3725061 -1.4766803
## ENSG00000273488 8.5844790 -0.10513006 0.3683837 -0.2853820
## ENSG00000273489 0.2758994 -0.06947899 0.1512520 -0.4593591
## ENSG00000273492 0.1059784 0.02314357 0.1512520 0.1530133
## ENSG00000273493 0.1061417 0.02314357 0.1512520 0.1530133
## pvalue padj
## <numeric> <numeric>
## ENSG00000000003 0.0001535423 0.001279829
## ENSG00000000419 0.0655868795 0.196015831
## ENSG00000000457 0.7937652416 0.912936622
## ENSG00000000460 0.7354072415 0.883203048
## ENSG00000000938 0.5782236287 NA
## ... ... ...
## ENSG00000273487 0.1397614 0.3376252
## ENSG00000273488 0.7753515 0.9032267
## ENSG00000273489 0.6459763 NA
## ENSG00000273492 0.8783878 NA
## ENSG00000273493 0.8783878 NAridx <- head(order(result$padj), 10)
top = result[ridx,]
top## log2 fold change (MAP): dex untrt vs trt
## Wald test p-value: dex untrt vs trt
## DataFrame with 10 rows and 6 columns
## baseMean log2FoldChange lfcSE stat
## <numeric> <numeric> <numeric> <numeric>
## ENSG00000152583 997.4398 -4.313962 0.17213733 -25.06116
## ENSG00000165995 495.0929 -3.186823 0.12815654 -24.86664
## ENSG00000101347 12703.3871 -3.618734 0.14894336 -24.29604
## ENSG00000120129 3409.0294 -2.871488 0.11824908 -24.28338
## ENSG00000189221 2341.7673 -3.230395 0.13667447 -23.63569
## ENSG00000211445 12285.6151 -3.553360 0.15798211 -22.49217
## ENSG00000157214 3009.2632 -1.948723 0.08867432 -21.97618
## ENSG00000162614 5393.1017 -2.003487 0.09269629 -21.61345
## ENSG00000125148 3656.2528 -2.167122 0.10354724 -20.92882
## ENSG00000154734 30315.1355 -2.286778 0.11308412 -20.22192
## pvalue padj
## <numeric> <numeric>
## ENSG00000152583 1.319237e-138 2.360906e-134
## ENSG00000165995 1.708565e-136 1.528824e-132
## ENSG00000101347 2.158637e-130 1.287699e-126
## ENSG00000120129 2.937247e-130 1.314124e-126
## ENSG00000189221 1.656535e-123 5.929070e-120
## ENSG00000211445 4.952260e-112 1.477094e-108
## ENSG00000157214 4.867315e-107 1.244364e-103
## ENSG00000162614 1.342345e-103 3.002826e-100
## ENSG00000125148 2.926277e-97 5.818739e-94
## ENSG00000154734 6.278709e-91 1.123638e-87Comprehension
There are many opportunities to place the statistical results into biological context. An initial step is to map the cryptic Ensembl gene identifiers used to label regions of interest to more famliar HGNC gene symbols. For this we use the org.Hs.eg.db package, an example of a Bioconductor ‘annotation’ package containing curated data derived from public-domain resources and updated semi-annually. The mapId() function maps between identifier types, in our case to SYMBOL gene ids from ENSEMBL ids. We add these to the top table results so that they can be processed together with the statistical results.
library(org.Hs.eg.db)
top$Symbol <- mapIds(org.Hs.eg.db, rownames(top), "SYMBOL", "ENSEMBL")## 'select()' returned 1:1 mapping between keys and columnstop## log2 fold change (MAP): dex untrt vs trt
## Wald test p-value: dex untrt vs trt
## DataFrame with 10 rows and 7 columns
## baseMean log2FoldChange lfcSE stat
## <numeric> <numeric> <numeric> <numeric>
## ENSG00000152583 997.4398 -4.313962 0.17213733 -25.06116
## ENSG00000165995 495.0929 -3.186823 0.12815654 -24.86664
## ENSG00000101347 12703.3871 -3.618734 0.14894336 -24.29604
## ENSG00000120129 3409.0294 -2.871488 0.11824908 -24.28338
## ENSG00000189221 2341.7673 -3.230395 0.13667447 -23.63569
## ENSG00000211445 12285.6151 -3.553360 0.15798211 -22.49217
## ENSG00000157214 3009.2632 -1.948723 0.08867432 -21.97618
## ENSG00000162614 5393.1017 -2.003487 0.09269629 -21.61345
## ENSG00000125148 3656.2528 -2.167122 0.10354724 -20.92882
## ENSG00000154734 30315.1355 -2.286778 0.11308412 -20.22192
## pvalue padj Symbol
## <numeric> <numeric> <character>
## ENSG00000152583 1.319237e-138 2.360906e-134 SPARCL1
## ENSG00000165995 1.708565e-136 1.528824e-132 CACNB2
## ENSG00000101347 2.158637e-130 1.287699e-126 SAMHD1
## ENSG00000120129 2.937247e-130 1.314124e-126 DUSP1
## ENSG00000189221 1.656535e-123 5.929070e-120 MAOA
## ENSG00000211445 4.952260e-112 1.477094e-108 GPX3
## ENSG00000157214 4.867315e-107 1.244364e-103 STEAP2
## ENSG00000162614 1.342345e-103 3.002826e-100 NEXN
## ENSG00000125148 2.926277e-97 5.818739e-94 MT2A
## ENSG00000154734 6.278709e-91 1.123638e-87 ADAMTS1Reproducibility
The calculations here are made more reproducible by reporting the version of software used in the analysis, as follows:
sessionInfo()## R version 3.3.1 Patched (2016-10-12 r71512)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 16.04.1 LTS
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=en_US.UTF-8 LC_COLLATE=en_US.UTF-8
## [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] parallel stats4 stats graphics grDevices utils datasets
## [8] methods base
##
## other attached packages:
## [1] org.Hs.eg.db_3.4.0 AnnotationDbi_1.35.5
## [3] DESeq2_1.13.16 SummarizedExperiment_1.3.82
## [5] Biobase_2.33.4 GenomicRanges_1.26.1
## [7] GenomeInfoDb_1.10.0 IRanges_2.8.0
## [9] S4Vectors_0.12.0 BiocGenerics_0.20.0
## [11] BiocInstaller_1.24.0
##
## loaded via a namespace (and not attached):
## [1] Rcpp_0.12.7 formatR_1.4 RColorBrewer_1.1-2
## [4] plyr_1.8.4 XVector_0.14.0 bitops_1.0-6
## [7] tools_3.3.1 zlibbioc_1.20.0 rpart_4.1-10
## [10] digest_0.6.10 RSQLite_1.0.0 annotate_1.51.1
## [13] evaluate_0.10 tibble_1.2 gtable_0.2.0
## [16] lattice_0.20-34 Matrix_1.2-7.1 DBI_0.5-1
## [19] yaml_2.1.13 gridExtra_2.2.1 genefilter_1.55.2
## [22] stringr_1.1.0 knitr_1.14 cluster_2.0.5
## [25] locfit_1.5-9.1 nnet_7.3-12 grid_3.3.1
## [28] data.table_1.9.6 XML_3.98-1.4 survival_2.39-5
## [31] BiocParallel_1.8.1 foreign_0.8-67 rmarkdown_1.1
## [34] latticeExtra_0.6-28 Formula_1.2-1 geneplotter_1.51.0
## [37] ggplot2_2.1.0 magrittr_1.5 Hmisc_3.17-4
## [40] scales_0.4.0 htmltools_0.3.5 splines_3.3.1
## [43] assertthat_0.1 xtable_1.8-2 colorspace_1.2-7
## [46] stringi_1.1.2 acepack_1.3-3.3 RCurl_1.95-4.8
## [49] munsell_0.4.3 chron_2.3-47