pareg 1.2.0
This vignette is an introduction to the usage of pareg
. It estimates pathway enrichment scores by regressing differential expression p-values of all genes considered in an experiment on their membership to a set of biological pathways. These scores are computed using a regularized generalized linear model with LASSO and network regularization terms. The network regularization term is based on a pathway similarity matrix (e.g., defined by Jaccard similarity) and thus classifies this method as a modular enrichment analysis tool (Huang, Sherman, and Lempicki 2009).
if (!require("BiocManager", quietly = TRUE)) {
install.packages("BiocManager")
}
BiocManager::install("pareg")
We start our analysis by loading the pareg
package and other required libraries.
library(ggraph)
library(tidyverse)
library(ComplexHeatmap)
library(enrichplot)
library(pareg)
set.seed(42)
For the sake of this introductory example, we generate a synthetic pathway database with a pronounced clustering of pathways.
group_num <- 2
pathways_from_group <- 10
gene_groups <- purrr::map(seq(1, group_num), function(group_idx) {
glue::glue("g{group_idx}_gene_{seq_len(15)}")
})
genes_bg <- paste0("bg_gene_", seq(1, 50))
df_terms <- purrr::imap_dfr(
gene_groups,
function(current_gene_list, gene_list_idx) {
purrr::map_dfr(seq_len(pathways_from_group), function(pathway_idx) {
data.frame(
term = paste0("g", gene_list_idx, "_term_", pathway_idx),
gene = c(
sample(current_gene_list, 10, replace = FALSE),
sample(genes_bg, 10, replace = FALSE)
)
)
})
}
)
df_terms %>%
sample_n(5)
## term gene
## 1 g1_term_9 g1_gene_12
## 2 g1_term_5 g1_gene_7
## 3 g2_term_2 g2_gene_2
## 4 g1_term_3 bg_gene_47
## 5 g1_term_8 g1_gene_1
Before starting the actual enrichment estimation, we compute pairwise pathway similarities with pareg
’s helper function.
mat_similarities <- compute_term_similarities(
df_terms,
similarity_function = jaccard
)
hist(mat_similarities, xlab = "Term similarity")
We can see a clear clustering of pathways.
Heatmap(
mat_similarities,
name = "Similarity",
col = circlize::colorRamp2(c(0, 1), c("white", "black"))
)
We then select a subset of pathways to be activated. In a performance evaluation, these would be considered to be true positives.
active_terms <- similarity_sample(mat_similarities, 5)
active_terms
## [1] "g2_term_6" "g2_term_3" "g2_term_3" "g2_term_2" "g2_term_8"
The genes contained in the union of active pathways are considered to be differentially expressed.
de_genes <- df_terms %>%
filter(term %in% active_terms) %>%
distinct(gene) %>%
pull(gene)
other_genes <- df_terms %>%
distinct(gene) %>%
pull(gene) %>%
setdiff(de_genes)
The p-values of genes considered to be differentially expressed are sampled from a Beta distribution centered at \(0\). The p-values for all other genes are drawn from a Uniform distribution.
df_study <- data.frame(
gene = c(de_genes, other_genes),
pvalue = c(rbeta(length(de_genes), 0.1, 1), rbeta(length(other_genes), 1, 1)),
in_study = c(
rep(TRUE, length(de_genes)),
rep(FALSE, length(other_genes))
)
)
table(
df_study$pvalue <= 0.05,
df_study$in_study, dnn = c("sig. p-value", "in study")
)
## in study
## sig. p-value FALSE TRUE
## FALSE 34 17
## TRUE 1 28
Finally, we compute pathway enrichment scores.
fit <- pareg(
df_study %>% select(gene, pvalue),
df_terms,
network_param = 1, term_network = mat_similarities
)
## Loaded Tensorflow version 2.7.0
The results can be exported to a dataframe for further processing…
fit %>%
as.data.frame() %>%
arrange(desc(abs(enrichment))) %>%
head() %>%
knitr::kable()
term | enrichment |
---|---|
g2_term_6 | -0.6765993 |
g2_term_3 | -0.6015774 |
g2_term_2 | -0.5822843 |
g2_term_4 | -0.4234689 |
g2_term_8 | -0.4132009 |
g1_term_2 | 0.3973174 |
…and also visualized in a pathway network view.
plot(fit, min_similarity = 0.1)
To provide a wider range of visualization options, the result can be transformed into an object which is understood by the functions of the enrichplot package.
obj <- as_enrichplot_object(fit)
dotplot(obj) +
scale_colour_continuous(name = "Enrichment Score")
## Scale for colour is already present.
## Adding another scale for colour, which will replace the existing scale.
treeplot(obj) +
scale_colour_continuous(name = "Enrichment Score")
## Scale for colour is already present.
## Adding another scale for colour, which will replace the existing scale.
sessionInfo()
## R version 4.2.2 (2022-10-31 ucrt)
## Platform: x86_64-w64-mingw32/x64 (64-bit)
## Running under: Windows Server x64 (build 20348)
##
## Matrix products: default
##
## locale:
## [1] LC_COLLATE=C
## [2] LC_CTYPE=English_United States.utf8
## [3] LC_MONETARY=English_United States.utf8
## [4] LC_NUMERIC=C
## [5] LC_TIME=English_United States.utf8
##
## attached base packages:
## [1] grid stats graphics grDevices utils datasets methods
## [8] base
##
## other attached packages:
## [1] pareg_1.2.0 tfprobability_0.15.1 tensorflow_2.9.0
## [4] enrichplot_1.18.0 ComplexHeatmap_2.14.0 forcats_0.5.2
## [7] stringr_1.4.1 dplyr_1.0.10 purrr_0.3.5
## [10] readr_2.1.3 tidyr_1.2.1 tibble_3.1.8
## [13] tidyverse_1.3.2 ggraph_2.1.0 ggplot2_3.4.0
## [16] BiocStyle_2.26.0
##
## loaded via a namespace (and not attached):
## [1] utf8_1.2.2 reticulate_1.26 tidyselect_1.2.0
## [4] RSQLite_2.2.18 AnnotationDbi_1.60.0 BiocParallel_1.32.1
## [7] scatterpie_0.1.8 munsell_0.5.0 codetools_0.2-18
## [10] future_1.29.0 withr_2.5.0 keras_2.9.0
## [13] colorspace_2.0-3 GOSemSim_2.24.0 Biobase_2.58.0
## [16] highr_0.9 knitr_1.40 stats4_4.2.2
## [19] DOSE_3.24.1 listenv_0.8.0 labeling_0.4.2
## [22] GenomeInfoDbData_1.2.9 matrixLaplacian_1.0 polyclip_1.10-4
## [25] bit64_4.0.5 farver_2.1.1 parallelly_1.32.1
## [28] vctrs_0.5.0 treeio_1.22.0 generics_0.1.3
## [31] xfun_0.34 timechange_0.1.1 R6_2.5.1
## [34] doParallel_1.0.17 GenomeInfoDb_1.34.3 clue_0.3-62
## [37] graphlayouts_0.8.3 bitops_1.0-7 cachem_1.0.6
## [40] fgsea_1.24.0 gridGraphics_0.5-1 assertthat_0.2.1
## [43] scales_1.2.1 googlesheets4_1.0.1 gtable_0.3.1
## [46] Cairo_1.6-0 globals_0.16.1 tidygraph_1.2.2
## [49] rlang_1.0.6 zeallot_0.1.0 scatterplot3d_0.3-42
## [52] GlobalOptions_0.1.2 splines_4.2.2 lazyeval_0.2.2
## [55] gargle_1.2.1 broom_1.0.1 BiocManager_1.30.19
## [58] yaml_2.3.6 reshape2_1.4.4 modelr_0.1.10
## [61] backports_1.4.1 qvalue_2.30.0 tools_4.2.2
## [64] bookdown_0.30 ggplotify_0.1.0 ellipsis_0.3.2
## [67] jquerylib_0.1.4 RColorBrewer_1.1-3 proxy_0.4-27
## [70] BiocGenerics_0.44.0 Rcpp_1.0.9 plyr_1.8.8
## [73] progress_1.2.2 base64enc_0.1-3 zlibbioc_1.44.0
## [76] RCurl_1.98-1.9 prettyunits_1.1.1 GetoptLong_1.0.5
## [79] viridis_0.6.2 cowplot_1.1.1 S4Vectors_0.36.0
## [82] haven_2.5.1 ggrepel_0.9.2 cluster_2.1.4
## [85] fs_1.5.2 furrr_0.3.1 magrittr_2.0.3
## [88] magick_2.7.3 data.table_1.14.4 circlize_0.4.15
## [91] reprex_2.0.2 googledrive_2.0.0 whisker_0.4
## [94] ggnewscale_0.4.8 matrixStats_0.62.0 hms_1.1.2
## [97] patchwork_1.1.2 evaluate_0.18 HDO.db_0.99.1
## [100] readxl_1.4.1 IRanges_2.32.0 gridExtra_2.3
## [103] shape_1.4.6 tfruns_1.5.1 compiler_4.2.2
## [106] crayon_1.5.2 shadowtext_0.1.2 htmltools_0.5.3
## [109] ggfun_0.0.8 tzdb_0.3.0 aplot_0.1.8
## [112] lubridate_1.9.0 DBI_1.1.3 tweenr_2.0.2
## [115] dbplyr_2.2.1 MASS_7.3-58.1 Matrix_1.5-3
## [118] cli_3.4.1 parallel_4.2.2 igraph_1.3.5
## [121] pkgconfig_2.0.3 xml2_1.3.3 foreach_1.5.2
## [124] ggtree_3.6.2 bslib_0.4.1 XVector_0.38.0
## [127] rvest_1.0.3 yulab.utils_0.0.5 digest_0.6.30
## [130] Biostrings_2.66.0 rmarkdown_2.18 cellranger_1.1.0
## [133] fastmatch_1.1-3 tidytree_0.4.1 nloptr_2.0.3
## [136] rjson_0.2.21 lifecycle_1.0.3 nlme_3.1-160
## [139] jsonlite_1.8.3 viridisLite_0.4.1 fansi_1.0.3
## [142] pillar_1.8.1 lattice_0.20-45 KEGGREST_1.38.0
## [145] fastmap_1.1.0 httr_1.4.4 GO.db_3.16.0
## [148] glue_1.6.2 png_0.1-7 iterators_1.0.14
## [151] bit_4.0.4 ggforce_0.4.1 stringi_1.7.8
## [154] sass_0.4.2 blob_1.2.3 memoise_2.0.1
## [157] ape_5.6-2
Huang, Da Wei, Brad T Sherman, and Richard A Lempicki. 2009. “Bioinformatics Enrichment Tools: Paths Toward the Comprehensive Functional Analysis of Large Gene Lists.” Nucleic Acids Research 37 (1): 1–13.