Last updated: 2019-01-12
workflowr checks: (Click a bullet for more information)Great! Since the R Markdown file has been committed to the Git repository, you know the exact version of the code that produced these results.
Great job! The global environment was empty. Objects defined in the global environment can affect the analysis in your R Markdown file in unknown ways. For reproduciblity it’s best to always run the code in an empty environment.
set.seed(20180714)
The command set.seed(20180714)
was run prior to running the code in the R Markdown file. Setting a seed ensures that any results that rely on randomness, e.g. subsampling or permutations, are reproducible.
Great job! Recording the operating system, R version, and package versions is critical for reproducibility.
wflow_publish
or wflow_git_commit
). workflowr only checks the R Markdown file, but you know if there are other scripts or data files that it depends on. Below is the status of the Git repository when the results were generated:
Ignored files:
Ignored: .DS_Store
Ignored: .Rhistory
Ignored: .Rproj.user/
Ignored: docs/.DS_Store
Ignored: docs/figure/.DS_Store
Untracked files:
Untracked: analysis/gd_notes.Rmd
Unstaged changes:
Modified: analysis/brain.Rmd
Note that any generated files, e.g. HTML, png, CSS, etc., are not included in this status report because it is ok for generated content to have uncommitted changes. File | Version | Author | Date | Message |
---|---|---|---|---|
Rmd | 7951c3c | Jason Willwerscheid | 2019-01-12 | workflowr::wflow_publish(“analysis/flashier_bench.Rmd”) |
First, I fit the “strong” subset of SNP-gene association statistics used in Urbut, Wang, Carbonetto, and Stephens (2018) (the strong.z
dataset found here). I fit five FLASH factors with normal-mixture priors using flashr
and using flashier
with backfit.order
set to "dropout"
, "sequential"
, and "montaigne"
.
The data is a dense 16k x 44 matrix that takes up 7.0 MB of memory when loaded into R
. I used the broadwl
partition of the midway2
RCC cluster with 4 CPUs and 8 GB of memory, and I used Gao Wang’s monitor_memory.py
script to test memory usage, as recommended in Peter Carbonetto’s large-scale data analysis tutorial.
VMS (GB) | RSS (GB) | Greedy (s/iter) | Backfit (s/iter) | Backfit iter | Obj diff | Time (min) | |
---|---|---|---|---|---|---|---|
flashr | 0.67 | 0.42 | 0.56 | 0.41 | 195 | 1894 | 2.22 |
dropout | 0.48 | 0.30 | 0.41 | 0.41 | 182 | 1894 | 1.98 |
sequential | 0.40 | 210 | 1894 | 2.15 | |||
montaigne | 0.41 | 324 | 1894 | 2.97 |
Next, I fit the droplet-based 3’ scRNA-seq dataset analyzed in Montoro et al. (2018) (the data can be obtained here). I performed a log-plus-one transform of the data, then I fit five FLASH factors using normal-mixture priors.
The data matrix is 18k x 7k and takes up 1011 MB of memory when loaded into R
as a dense matrix. However, only 9.3% of entries are nonzero, so the data can also be loaded as a sparse Matrix
object, in which case the data takes up 143 MB of memory. I fit FLASH objects to the larger matrix
object using the same four approaches used to fit the GTEx dataset, then I fit a FLASH object to the sparse Matrix
object using flashier
with backfit.order = "dropout"
(flashr
does not support objects of class Matrix
). All fits were performed on the broadwl
partition of the midway2
RCC cluster using 4 CPUs and 32 GB of memory.
VMS (GB) | RSS (GB) | Greedy (s/iter) | Backfit (s/iter) | Backfit iter | Obj diff | Time (min) | |
---|---|---|---|---|---|---|---|
flashr | 18.4 | 18.1 | 23.49 | 15.64 | 250 | 27002 | 101.97 |
dropout | 3.9 | 3.7 | 1.91 | 1.92 | 285 | 26982 | 11.27 |
sequential | 2.71 | 250 | 27109 | 13.45 | |||
montaigne | 1.82 | 100 | 13004 | 5.20 | |||
sparse | 1.1 | 0.9 | 0.90 | 0.88 | 285 | 26982 | 5.22 |
Finally, I fit the larger full-length scRNA “PulseSeq” dataset from Montoro et al. (2018). The dataset is about ten times larger than the droplet-based scRNA-seq dataset, so it was not feasible to use flashr
. I again fit five factors using normal-mixture priors, and I set backfit.order = "dropout"
.
The dataset is 22k x 66k, with 9.3% of entries not equal to zero, and occupies 1.49 GB of memory when loaded into R
as a sparse Matrix
object. (I did not attempt to fit a larger matrix
object.) The fit was again performed on broadwl
using 4 CPUs and 32 GB of memory.
VMS (GB) | RSS (GB) | Greedy (s/iter) | Backfit (s/iter) | Backfit iter | Obj diff | Time (min) | |
---|---|---|---|---|---|---|---|
sparse | 6 | 5.8 | 7.74 | 3.9 | 334 | 300023 | 27.23 |
sessionInfo()
R version 3.4.3 (2017-11-30)
Platform: x86_64-apple-darwin15.6.0 (64-bit)
Running under: macOS High Sierra 10.13.6
Matrix products: default
BLAS: /Library/Frameworks/R.framework/Versions/3.4/Resources/lib/libRblas.0.dylib
LAPACK: /Library/Frameworks/R.framework/Versions/3.4/Resources/lib/libRlapack.dylib
locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
attached base packages:
[1] stats graphics grDevices utils datasets methods base
loaded via a namespace (and not attached):
[1] workflowr_1.0.1 Rcpp_1.0.0 digest_0.6.18
[4] rprojroot_1.3-2 R.methodsS3_1.7.1 backports_1.1.2
[7] magrittr_1.5 git2r_0.21.0 evaluate_0.12
[10] highr_0.7 stringi_1.2.4 whisker_0.3-2
[13] R.oo_1.21.0 R.utils_2.6.0 rmarkdown_1.10
[16] tools_3.4.3 stringr_1.3.1 xfun_0.4
[19] yaml_2.2.0 compiler_3.4.3 htmltools_0.3.6
[22] knitr_1.20.22
This reproducible R Markdown analysis was created with workflowr 1.0.1