compcodeR 1.32.1
library(compcodeR)
The compcodeR R package can generate RNAseq counts data and compare the relative performances of various popular differential analysis detection tools (Soneson and Delorenzi (2013)).
Using the same framework, this document shows how to generate “orthologous gene” (OG) expression for different species, taking into account their varying lengths, and their phylogenetic relationships, as encoded by an evolutionary tree.
This vignette provides a tutorial on how to use the “phylogenetic” functionalities of compcodeR.
It assumes that the reader is already familiar with the compcodeR
package vignette.
phyloCompData
classThe phyloCompData
class extends the compData
class
of the compcodeR package
to account for phylogeny and length information needed in the representation of
OG expression data.
A phyloCompData
object contains all the slots of a
compData
object,
with an added slot containing a phylogenetic tree
with ape
format phylo
,
and a length matrix.
It can also contain some added variable information, such as species names.
More detailed information about the phyloCompData
class are available in the
section on the phylo data object.
After conducting a differential expression analysis, the phyloCompData
object
has the same added information than the compData
object
(see the result object
in the compcodeR package vignette).
The workflow for working with the inter-species extension is very similar to the already existing workflow of the compcodeR package. In this section, we recall this workflow, stressing out the added functionalities.
The simulations are performed following the description by Bastide et al. (2022).
We use here the phylogenetic tree issued from Stern et al. (2017), normalized to unit height, that has \(14\) species with up to 3 replicates, for a total number of sample equal to \(34\) (see Figure below).
library(ape)
tree <- system.file("extdata", "Stern2018.tree", package = "compcodeR")
tree <- read.tree(tree)
Note that any other tree could be used, for instance randomly generated
using a birth-death process, see e.g. function rphylo
in the
ape
package.
To conduct a differential analysis, each species must be attributed a condition. Because of the phylogenetic structure, the condition design does matter, and have a strong influence on the data produced. Here, we assume that the conditions are mapped on the tree in a balanced way (“alt” design), which is the “best case scenario”.
# link each sample to a species
id_species <- factor(sub("_.*", "", tree$tip.label))
names(id_species) <- tree$tip.label
# Assign a condition to each species
species_names <- unique(id_species)
species_names[c(length(species_names)-1, length(species_names))] <- species_names[c(length(species_names), length(species_names)-1)]
cond_species <- rep(c(1, 2), length(species_names) / 2)
names(cond_species) <- species_names
# map them on the tree
id_cond <- id_species
id_cond <- cond_species[as.vector(id_cond)]
id_cond <- as.factor(id_cond)
names(id_cond) <- tree$tip.label
We can plot the assigned conditions on the tree to visualize them.
plot(tree, label.offset = 0.01)
tiplabels(pch = 19, col = c("#D55E00", "#009E73")[id_cond])
Using this tree with associated condition design, we can then generate a dataset
using a “phylogenetic Poisson Log Normal” (pPLN) distribution.
We use here a Brownian Motion (BM) model of evolution for the latent phylogenetic
log normal continuous trait, and assume that the phylogenetic model accounts for
\(90\%\) of the latent trait variance
(i.e. there is an added uniform intra-species variance representing \(10\%\) of the
total latent trait variation).
Using the "auto"
setup, the counts are simulated so that they match empirical
moments found in Stern and Crandall (2018).
OG lengths are also drawn from a pPLN model, so that their moments match those
of the empirical dataset of Stern and Crandall (2018).
We choose to simulate \(2000\) OGs, \(10\%\) of which are differentially expressed, with an effect size of \(3\).
The following code creates a phyloCompData
object containing the simulated
data set and saves it to a file named "alt_BM_repl1.rds"
.
set.seed(12890926)
alt_BM <- generateSyntheticData(dataset = "alt_BM",
n.vars = 2000, samples.per.cond = 17,
n.diffexp = 200, repl.id = 1,
seqdepth = 1e7, effect.size = 3,
fraction.upregulated = 0.5,
output.file = "alt_BM_repl1.rds",
## Phylogenetic parameters
tree = tree, ## Phylogenetic tree
id.species = id_species, ## Species structure of samples
id.condition = id_cond, ## Condition design
model.process = "BM", ## The latent trait follows a BM
prop.var.tree = 0.9, ## Tree accounts for 90% of the variance
lengths.relmeans = "auto", ## OG length mean and dispersion
lengths.dispersions = "auto") ## are taken from an empirical exemple
The summarizeSyntheticDataSet
works the same way as in the base
compcodeR
package, generating a report that summarize
all the parameters used in the simulation, and showing some diagnostic plots.
summarizeSyntheticDataSet(data.set = "alt_BM_repl1.rds",
output.filename = "alt_BM_repl1_datacheck.html")
When applied to a phyloCompData
object,
it provides some extra diagnostics, related to the phylogenetic nature of the data.
In particular, it contains MA-plots with TPM-normalized expression levels to
take OG length into account, which generally makes the original signal
clearer.
It also shows a log2 normalized counts heatmap plotted along the phylogeny, illustrating the phylogenetic structure of the differentially expressed OGs.
Differential expression analysis can be conducted using the same framework
used in the compcodeR
package,
through the runDiffExp
function.
All the standard methods can be used. To account for the phylogenetic nature of the data and for the varying length of the OGs, some methods have been added to the pool.
The code below applies three differential expression methods to the data set generated above:
the DESeq2 method adapted for varying lengths,
the log2(TPM)
transformation for length normalization,
combined with limma, using the trend
empirical Bayes correction,
and accounting for species-related correlations, and
the phylogenetic regression tool
phylolm
applied on the same log2(TPM)
.
runDiffExp(data.file = "alt_BM_repl1.rds",
result.extent = "DESeq2", Rmdfunction = "DESeq2.createRmd",
output.directory = ".",
fit.type = "parametric", test = "Wald")
runDiffExp(data.file = "alt_BM_repl1.rds",
result.extent = "lengthNorm.limma", Rmdfunction = "lengthNorm.limma.createRmd",
output.directory = ".",
norm.method = "TMM",
length.normalization = "TPM",
data.transformation = "log2",
trend = FALSE, block.factor = "id.species")
runDiffExp(data.file = "alt_BM_repl1.rds",
result.extent = "phylolm", Rmdfunction = "phylolm.createRmd",
output.directory = ".",
norm.method = "TMM",
model = "BM", measurement_error = TRUE,
extra.design.covariates = NULL,
length.normalization = "TPM",
data.transformation = "log2")
As for a regular compcodeR analysis,
example calls are provided in the reference manual (see the help pages for the runDiffExp
function),
and a list of all available methods can be obtained with the listcreateRmd()
function.
listcreateRmd()
#> [1] "DESeq2.createRmd"
#> [2] "DESeq2.length.createRmd"
#> [3] "DSS.createRmd"
#> [4] "EBSeq.createRmd"
#> [5] "NBPSeq.createRmd"
#> [6] "NOISeq.prenorm.createRmd"
#> [7] "TCC.createRmd"
#> [8] "baySeq.createRmd"
#> [9] "edgeR.GLM.createRmd"
#> [10] "edgeR.exact.createRmd"
#> [11] "lengthNorm.limma.createRmd"
#> [12] "logcpm.limma.createRmd"
#> [13] "phylolm.createRmd"
#> [14] "sqrtcpm.limma.createRmd"
#> [15] "ttest.createRmd"
#> [16] "voom.limma.createRmd"
#> [17] "voom.ttest.createRmd"
Given that the phyloCompData
object has the same structure with respect to the
slots added by the differential expression analysis
(see the result object,
the procedure to compare results from several differential expression methods
is exactly the same as for a compData
object, and can be found in the
corresponding section
section of the compcodeR vignette.
As for a compData
object,
it is still possible to input user-defined data to produce a phyloCompData
object for differential expression methods comparisons.
One only needs to provide the additional information needed, that is
the phylogenetic tree, and the length matrix.
The constructor method will make sure that the tree is consistent with the count
and length matrices, with the same dimensions and consistent species names.
## Phylogentic tree with replicates
tree <- read.tree(text = "(((A1:0,A2:0,A3:0):1,B1:1):1,((C1:0,C2:0):1.5,(D1:0,D2:0):1.5):0.5);")
## Sample annotations
sample.annotations <- data.frame(
condition = c(1, 1, 1, 1, 2, 2, 2, 2), # Condition of each sample
id.species = c("A", "A", "A", "B", "C", "C", "D", "D") # Species of each sample
)
## Count Matrix
count.matrix <- round(matrix(1000*runif(8000), 1000))
## Length Matrix
length.matrix <- round(matrix(1000*runif(8000), 1000))
## Names must match
colnames(count.matrix) <- colnames(length.matrix) <- rownames(sample.annotations) <- tree$tip.label
## Extra infos
info.parameters <- list(dataset = "mydata", uID = "123456")
## Creation of the object
cpd <- phyloCompData(count.matrix = count.matrix,
sample.annotations = sample.annotations,
info.parameters = info.parameters,
tree = tree,
length.matrix = length.matrix)
## Check
check_phyloCompData(cpd)
#> [1] TRUE
To use your own differential expression code,
you can follow the
base compcodeR
instructions
in the compcodeR vignette.
The phylocompData
data object is an S4 object that extends
the compData
object,
with the following added slots:
tree
[class phylo
] (mandatory) – the phylogenetic tree describing the relationships between samples.length.matrix
[class matrix
] (mandatory) – the OG length matrix, with rows representing genes and columns representing samples.
generateSyntheticData
, the sample.annotations
data frame has added column:
id.species
[class character
or numeric
] – the species for each sample.
Should match with the tip.label
of the tree
slot.When produced with generateSyntheticData
, the variable.annotations
data frame has an added columns:
lengths.relmeans
[class numeric
] – the true mean values used in the simulations of the OG lengths.lengths.dispersions
[class numeric
] – the true dispersion values used in the simulations of the OG lengths.M.value.TPM
[class numeric
] – the estimated log2-fold change between conditions 1 and 2 for each OG using TPM length normalization.A.value.TPM
[class numeric
] – the estimated average expression in conditions 1 and 2 for each OG using TPM length normalization.prop.var.tree
[class numeric
] – the proportion of the variance explained by the phylogeny for each gene.The same way as the compData
object, the phyloCompData
object needs to be saved to a file with extension .rds
.
The evaluation metrics are unchanged, and described in the corresponding section section of the compcodeR vignette.
sessionInfo()
#> R version 4.2.1 (2022-06-23)
#> Platform: x86_64-pc-linux-gnu (64-bit)
#> Running under: Ubuntu 20.04.5 LTS
#>
#> Matrix products: default
#> BLAS: /home/biocbuild/bbs-3.15-bioc/R/lib/libRblas.so
#> LAPACK: /home/biocbuild/bbs-3.15-bioc/R/lib/libRlapack.so
#>
#> locale:
#> [1] LC_CTYPE=en_US.UTF-8
#> [2] LC_NUMERIC=C
#> [3] LC_TIME=en_GB
#> [4] LC_COLLATE=C
#> [5] LC_MONETARY=en_US.UTF-8
#> [6] LC_MESSAGES=en_US.UTF-8
#> [7] LC_PAPER=en_US.UTF-8
#> [8] LC_NAME=C
#> [9] LC_ADDRESS=C
#> [10] LC_TELEPHONE=C
#> [11] LC_MEASUREMENT=en_US.UTF-8
#> [12] LC_IDENTIFICATION=C
#>
#> attached base packages:
#> [1] stats graphics grDevices utils datasets
#> [6] methods base
#>
#> other attached packages:
#> [1] ape_5.6-2 compcodeR_1.32.1 sm_2.2-5.7.1
#> [4] BiocStyle_2.24.0
#>
#> loaded via a namespace (and not attached):
#> [1] Rcpp_1.0.9 locfit_1.5-9.6
#> [3] lattice_0.20-45 zoo_1.8-11
#> [5] gtools_3.9.3 assertthat_0.2.1
#> [7] digest_0.6.29 utf8_1.2.2
#> [9] R6_2.5.1 fBasics_4021.92
#> [11] evaluate_0.16 highr_0.9
#> [13] ggplot2_3.3.6 pillar_1.8.1
#> [15] gplots_3.1.3 rlang_1.0.5
#> [17] spatial_7.3-15 jquerylib_0.1.4
#> [19] magick_2.7.3 rpart_4.1.16
#> [21] rmarkdown_2.16 stringr_1.4.1
#> [23] munsell_0.5.0 compiler_4.2.1
#> [25] xfun_0.33 pkgconfig_2.0.3
#> [27] stable_1.1.6 tcltk_4.2.1
#> [29] htmltools_0.5.3 tidyselect_1.1.2
#> [31] tibble_3.1.8 bookdown_0.29
#> [33] edgeR_3.38.4 stabledist_0.7-1
#> [35] fansi_1.0.3 dplyr_1.0.10
#> [37] MASS_7.3-58.1 bitops_1.0-7
#> [39] grid_4.2.1 nlme_3.1-159
#> [41] jsonlite_1.8.0 gtable_0.3.1
#> [43] lifecycle_1.0.2 DBI_1.1.3
#> [45] magrittr_2.0.3 scales_1.2.1
#> [47] KernSmooth_2.23-20 statip_0.2.3
#> [49] cli_3.4.0 stringi_1.7.8
#> [51] cachem_1.0.6 ROCR_1.0-11
#> [53] rmutil_1.1.9 limma_3.52.3
#> [55] timeDate_4021.104 bslib_0.4.0
#> [57] generics_0.1.3 vctrs_0.4.1
#> [59] vioplot_0.3.7 modeest_2.4.0
#> [61] tools_4.2.1 glue_1.6.2
#> [63] markdown_1.1 purrr_0.3.4
#> [65] parallel_4.2.1 fastmap_1.1.0
#> [67] yaml_2.3.5 clue_0.3-61
#> [69] colorspace_2.0-3 cluster_2.1.4
#> [71] BiocManager_1.30.18 caTools_1.18.2
#> [73] timeSeries_4021.104 knitr_1.40
#> [75] sass_0.4.2
Bastide, Paul, Charlotte Soneson, Olivier Lespinet, and Mélina Gallopin. 2022. “Benchmark of Differential Gene Expression Analysis Methods for Inter-Species Rna-Seq Data Using a Phylogenetic Simulation Framework.” bioRxiv Preprint. https://doi.org/10.1101/2022.01.21.476612.
Soneson, Charlotte, and Mauro Delorenzi. 2013. “A Comparison of Methods for Differential Expression Analysis of RNA-seq Data.” BMC Bioinformatics 14: 91.
Stern, David B., Jesse Breinholt, Carlos Pedraza-Lara, Marilú López-Mejía, Christopher L. Owen, Heather Bracken-Grissom, James W. Fetzner, and Keith A. Crandall. 2017. “Phylogenetic Evidence from Freshwater Crayfishes That Cave Adaptation Is Not an Evolutionary Dead-End.” Evolution 71 (10): 2522–32. https://doi.org/10.1111/evo.13326.
Stern, David B., and Keith A. Crandall. 2018. “The Evolution of Gene Expression Underlying Vision Loss in Cave Animals.” Molecular Biology and Evolution 35 (8): 2005–14. https://doi.org/10.1093/molbev/msy106.