在R markdown文档中将sessionInfo()放在末尾几乎已经成为一种标准。sessionInfo()能够打印出加载到R session中的包的列表。但这些包之间的依赖关系又如何呢？在这篇博文中，让我们来看看。
让我们打开一个新的R session并仅加载ggplot2包：
library(ggplot2)
接下来我们运行sessionInfo()：
x1 = sessionInfo()
打印x1：
x1
## R version 4.3.1 (2023-06-16)
## Platform: x86_64-apple-darwin20 (64-bit)
## Running under: macOS Ventura 13.2.1
##
## Matrix products: default
## BLAS: /Library/Frameworks/R.framework/Versions/4.3-x86_64/Resources/lib/libRblas.0.dylib
## LAPACK: /Library/Frameworks/R.framework/Versions/4.3-x86_64/Resources/lib/libRlapack.dylib; LAPACK version 3.11.0
##
## locale:
## [1] C/UTF-8/C/C/C/C
##
## time zone: Europe/Berlin
## tzcode source: internal
##
## attached base packages:
## [1] stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] ggplot2_3.4.4
##
## loaded via a namespace (and not attached):
## [1] utf8_1.2.3 R6_2.5.1 tidyselect_1.2.0 magrittr_2.0.3 gtable_0.3.4
## [6] glue_1.6.2 tibble_3.2.1 pkgconfig_2.0.3 generics_0.1.3 dplyr_1.1.3
## [11] lifecycle_1.0.3 cli_3.6.1 fansi_1.0.5 scales_1.2.1 grid_4.3.1
## [16] vctrs_0.6.4 withr_2.5.1 compiler_4.3.1 munsell_0.5.0 pillar_1.9.0
## [21] colorspace_2.1-0 rlang_1.1.1
只有二十几个包被载入，看起来不是很多。返回变量x1是一个列表，其中包含了直接或间接加载的包列表。这些包可以分为三组：
- 基础包：R附带的基础包，例如grid、methods。
- 其他包：例如通过
library()加载的包。它们在搜索路径上可见（通过search()函数可以看到完整的搜索路径）。 - 加载包：通过“其他包”加载到R session中，但在搜索路径上不可见。
base_pkgs = x1$basePkgs
other_pkgs = sapply(x1$otherPkgs, function(x) x$Package)
loaded_pkgs = sapply(x1$loadedOnly, function(x) x$Package)
对于session中的包，现在我们需要知道它们的依赖关系。这里我们使用pkgndep包。所有本地安装的包都作为“包数据库”用来来查询依赖关系。
library(pkgndep)
db = reformat_db(installed.packages())
## prepare dependency table...
## prepare reverse dependency table...
loaded_pkgs均通过other_packages以直接或者间接的方式导入R session，所以
我们遍历每个other_packages中的包，获取其直接和间接的上游依赖关系。这里我们只使用“强”依赖，即依赖关系为“Depends”、“Imports”和“LinkingTo”的依赖包。
mat = matrix(nrow = 0, ncol = 3)
for(pkg in other_pkgs) {
mat = rbind(mat, db$package_dependencies(pkg, recursive = TRUE, which = "strong"))
}
mat = unique(mat)
基础包直接随着R发布，我们不关心基于它们的依赖关系，因此我们删除与基础包的依赖关系。我们只将包限制为“其他包”和“加载包”。
all_pkgs = c(other_pkgs, loaded_pkgs)
mat = mat[mat[, 1] %in% all_pkgs & mat[, 2] %in% all_pkgs, , drop = FALSE]
mat = mat[!mat[, 1] %in% pkgndep:::BASE_PKGS | mat[, 2] %in% pkgndep:::BASE_PKGS, , drop = FALSE]
head(mat)
## package dependency dep_fields
## [1,] "ggplot2" "cli" "Imports"
## [2,] "ggplot2" "glue" "Imports"
## [3,] "ggplot2" "grid" "Imports"
## [4,] "ggplot2" "gtable" "Imports"
## [5,] "ggplot2" "lifecycle" "Imports"
## [6,] "ggplot2" "rlang" "Imports"
接下来我们使用DiagrammeR包来可视化依赖关系。我们首先生成DOT代码：
all_nodes = unique(c(mat[, 1], mat[, 2], other_pkgs, loaded_pkgs))
node_col = rep("black", length(all_nodes))
node_col[all_nodes %in% other_pkgs] = "red"
node_col[all_nodes %in% loaded_pkgs] = "blue"
library(glue)
nodes = glue(" \"{all_nodes}\" [color=\"{node_col}\"];", collapse = FALSE)
dep_col = c(2, 4, 3, 5, 6)
dep_col = rgb(t(col2rgb(dep_col)), max = 255)
names(dep_col) = c("Depends", "Imports", "LinkingTo", "Suggests", "Enhances")
edges = glue(" \"{mat[, 1]}\" -> \"{mat[, 2]}\" [color=\"{dep_col[mat[, 3]]}\"];", collapse = FALSE)
dot = paste(
c("digraph {",
" nodesep=0.05",
" rankdir=LR;",
" graph [overlap = true];",
" node[shape = box];",
nodes,
edges,
"}"),
collapse = "\n"
)
cat(dot)
## digraph {
## nodesep=0.05
## rankdir=LR;
## graph [overlap = true];
## node[shape = box];
## "ggplot2" [color="red"];
## "gtable" [color="blue"];
## "lifecycle" [color="blue"];
## "scales" [color="blue"];
## "tibble" [color="blue"];
## "vctrs" [color="blue"];
## "munsell" [color="blue"];
## "pillar" [color="blue"];
## "cli" [color="blue"];
## "glue" [color="blue"];
## "grid" [color="blue"];
## "rlang" [color="blue"];
## "withr" [color="blue"];
## "R6" [color="blue"];
## "fansi" [color="blue"];
## "magrittr" [color="blue"];
## "pkgconfig" [color="blue"];
## "colorspace" [color="blue"];
## "utf8" [color="blue"];
## "tidyselect" [color="blue"];
## "generics" [color="blue"];
## "dplyr" [color="blue"];
## "compiler" [color="blue"];
## "ggplot2" -> "cli" [color="#2297E6"];
## "ggplot2" -> "glue" [color="#2297E6"];
## "ggplot2" -> "grid" [color="#2297E6"];
## "ggplot2" -> "gtable" [color="#2297E6"];
## "ggplot2" -> "lifecycle" [color="#2297E6"];
## "ggplot2" -> "rlang" [color="#2297E6"];
## "ggplot2" -> "scales" [color="#2297E6"];
## "ggplot2" -> "tibble" [color="#2297E6"];
## "ggplot2" -> "vctrs" [color="#2297E6"];
## "ggplot2" -> "withr" [color="#2297E6"];
## "gtable" -> "cli" [color="#2297E6"];
## "gtable" -> "glue" [color="#2297E6"];
## "gtable" -> "grid" [color="#2297E6"];
## "gtable" -> "lifecycle" [color="#2297E6"];
## "gtable" -> "rlang" [color="#2297E6"];
## "lifecycle" -> "cli" [color="#2297E6"];
## "lifecycle" -> "glue" [color="#2297E6"];
## "lifecycle" -> "rlang" [color="#2297E6"];
## "scales" -> "cli" [color="#2297E6"];
## "scales" -> "glue" [color="#2297E6"];
## "scales" -> "lifecycle" [color="#2297E6"];
## "scales" -> "munsell" [color="#2297E6"];
## "scales" -> "R6" [color="#2297E6"];
## "scales" -> "rlang" [color="#2297E6"];
## "tibble" -> "fansi" [color="#2297E6"];
## "tibble" -> "lifecycle" [color="#2297E6"];
## "tibble" -> "magrittr" [color="#2297E6"];
## "tibble" -> "pillar" [color="#2297E6"];
## "tibble" -> "pkgconfig" [color="#2297E6"];
## "tibble" -> "rlang" [color="#2297E6"];
## "tibble" -> "vctrs" [color="#2297E6"];
## "vctrs" -> "cli" [color="#2297E6"];
## "vctrs" -> "glue" [color="#2297E6"];
## "vctrs" -> "lifecycle" [color="#2297E6"];
## "vctrs" -> "rlang" [color="#2297E6"];
## "munsell" -> "colorspace" [color="#2297E6"];
## "pillar" -> "cli" [color="#2297E6"];
## "pillar" -> "glue" [color="#2297E6"];
## "pillar" -> "lifecycle" [color="#2297E6"];
## "pillar" -> "rlang" [color="#2297E6"];
## "pillar" -> "utf8" [color="#2297E6"];
## "pillar" -> "vctrs" [color="#2297E6"];
## }
然后我们使用grViz()生成依赖关系网络图。
DiagrammeR::grViz(dot)
看起来依赖关系网络比简单列出包要略微复杂一些。图中还有一些未连接到ggplot2的包，例如dplyr。它们由ggplot2或其上游包作为“弱依赖”间接加载到R session中。
我们将代码包装为可以重复使用的函数loaded_pkgs()。在函数内部，我们使用callr包在一个新的R session中调用library(pkg)和获取session信息。
loaded_pkgs = function(pkg) {
for(i in seq_along(pkg)) {
library(pkg[i], character.only=TRUE)
}
session_info = sessionInfo()
base_pkgs = session_info$basePkgs
other_pkgs = sapply(session_info$otherPkgs, function(x)x$Package)
loaded_pkgs = sapply(session_info$loadedOnly, function(x)x$Package)
lt = list(base_pkgs = base_pkgs,
other_pkgs = other_pkgs,
loaded_pkgs = loaded_pkgs)
jsonlite::toJSON(lt)
}
dep_in_session = function(pkg, db, dep_group = "strong", rankdir = "LR") {
session_info = jsonlite::fromJSON(callr::r(loaded_pkgs, args = list(pkg = pkg)))
base_pkgs = session_info$base_pkgs
other_pkgs = session_info$other_pkgs
loaded_pkgs = session_info$loaded_pkgs
mat = matrix(nrow = 0, ncol = 3)
for(pkg in other_pkgs) {
mat = rbind(mat, db$package_dependencies(pkg, recursive = TRUE, which = dep_group))
}
mat = unique(mat)
mat = mat[!mat[, 1] %in% pkgndep:::BASE_PKGS | mat[, 2] %in% pkgndep:::BASE_PKGS, , drop = FALSE]
all_pkgs = c(other_pkgs, loaded_pkgs)
mat = mat[mat[, 1] %in% all_pkgs & mat[, 2] %in% all_pkgs, , drop = FALSE]
all_nodes = unique(c(mat[, 1], mat[, 2], other_pkgs, loaded_pkgs))
node_col = rep("black", length(all_nodes))
node_col[all_nodes %in% other_pkgs] = "red"
node_col[all_nodes %in% loaded_pkgs] = "blue"
nodes = glue::glue(" \"{all_nodes}\" [color=\"{node_col}\"];", collapse = FALSE)
dep_col = c(2, 4, 3, 5, 6)
dep_col = rgb(t(col2rgb(dep_col)), max = 255)
names(dep_col) = c("Depends", "Imports", "LinkingTo", "Suggests", "Enhances")
edges = glue::glue(" \"{mat[, 1]}\" -> \"{mat[, 2]}\" [color=\"{dep_col[mat[, 3]]}\"];", collapse = FALSE)
dot = paste(
c("digraph {",
" nodesep=0.05",
glue::glue(" rankdir={rankdir};"),
" graph [overlap = true];",
" node[shape = box];",
nodes,
edges,
"}"),
collapse = "\n"
)
DiagrammeR::grViz(dot)
}
在前面的示例中，我们只考虑了ggplot2上游的“强依赖关系”。如果我们包含所有依赖关系会怎样？
dep_in_session("ggplot2", db = db, dep_group = "all")
它变得更加复杂！特别是我们可以看到存在许多双向依赖关系，例如， A <-> B，其中A是B的强依赖项，而B是A的弱依赖项。
接下来让我们检查一个Bioconductor包DESeq2。
我们首先只考虑DESeq2上游的强依赖性。
dep_in_session("DESeq2", db = db, dep_group = "strong")
它已经非常复杂，但有趣的是，DESeq2的依赖项可以很好地分为两组。第一组与Bioconductor包相关，其中很多直接被附加到搜索路径（红色框：在搜索路径中可见；红色链接：“Depends”关系）；第二组主要与ggplot2及其上游包相关，它们都间接加载到R会话中（蓝色链接：“Imports”关系）。
如果我们包含所有依赖类型会怎样？现在我们需要将布局更改为“TB”样式（top-bottom）。正如我们所看到的，一个简单的命令library(DESeq2)给你的 R session带来了一个dependency monster。
dep_in_session("DESeq2", db = db, dep_group = "all", rankdir = "TB")
最后，让我们检查一下如果仅载入Seurat包。
dep_in_session("Seurat", db = db, dep_group = "strong")
眼花缭乱！
如果包含Seurat所有的依赖关系，会变得异常庞大和复杂。
dep_in_session("Seurat", db = db, dep_group = "all")
sessionInfo()
## R version 4.4.2 (2024-10-31)
## Platform: aarch64-apple-darwin20
## Running under: macOS 26.1
##
## Matrix products: default
## BLAS: /Library/Frameworks/R.framework/Versions/4.4-arm64/Resources/lib/libRblas.0.dylib
## LAPACK: /Library/Frameworks/R.framework/Versions/4.4-arm64/Resources/lib/libRlapack.dylib; LAPACK version 3.12.0
##
## locale:
## [1] C.UTF-8/UTF-8/C.UTF-8/C/C.UTF-8/C.UTF-8
##
## time zone: Europe/Berlin
## tzcode source: internal
##
## attached base packages:
## [1] stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] glue_1.8.0 pkgndep_1.99.3 knitr_1.50 colorout_1.3-2
##
## loaded via a namespace (and not attached):
## [1] DiagrammeR_1.0.11 sass_0.4.9 BiocVersion_3.20.0 shape_1.4.6.1
## [5] blogdown_1.19 digest_0.6.37 magrittr_2.0.3 evaluate_1.0.3
## [9] grid_4.4.2 RColorBrewer_1.1-3 bookdown_0.44 iterators_1.0.14
## [13] circlize_0.4.16 fastmap_1.2.0 foreach_1.5.2 doParallel_1.0.17
## [17] jsonlite_1.9.0 processx_3.8.6 GlobalOptions_0.1.4 ps_1.9.0
## [21] ComplexHeatmap_2.25.2 codetools_0.2-20 jquerylib_0.1.4 cli_3.6.4
## [25] rlang_1.1.5 crayon_1.5.3 visNetwork_2.1.2 cachem_1.1.0
## [29] yaml_2.3.10 tools_4.4.2 parallel_4.4.2 colorspace_2.1-1
## [33] GetoptLong_1.0.5 BiocGenerics_0.52.0 hash_2.2.6.3 R6_2.6.1
## [37] png_0.1-8 matrixStats_1.5.0 stats4_4.4.2 lifecycle_1.0.4
## [41] S4Vectors_0.44.0 htmlwidgets_1.6.4 IRanges_2.40.1 clue_0.3-66
## [45] cluster_2.1.6 callr_3.7.6 bslib_0.9.0 xfun_0.51
## [49] rstudioapi_0.17.1 rjson_0.2.23 htmltools_0.5.8.1 rmarkdown_2.29
## [53] compiler_4.4.2