1 Basics

1.1 Install derfinderHelper

R is an open-source statistical environment which can be easily modified to enhance its functionality via packages. derfinderHelper is a R package available via the Bioconductor repository for packages. R can be installed on any operating system from CRAN after which you can install derfinderHelper by using the following commands in your R session:

install.packages("BiocManager")
BiocManager::install("derfinderHelper")

## Check that you have a valid Bioconductor installation
BiocManager::valid()

1.2 Required knowledge

derfinderHelper is based on many other packages and in particular in those that have implemented the infrastructure needed for dealing with RNA-seq data. A derfinderHelper user is not expected to deal with those packages directly but will need to be familiar with derfinder.

If you are asking yourself the question “Where do I start using Bioconductor?” you might be interested in this blog post.

1.3 Asking for help

As package developers, we try to explain clearly how to use our packages and in which order to use the functions. But R and Bioconductor have a steep learning curve so it is critical to learn where to ask for help. The blog post quoted above mentions some but we would like to highlight the Bioconductor support site as the main resource for getting help: remember to use the derfinder or derfinderHelper tags and check the older posts. Other alternatives are available such as creating GitHub issues and tweeting. However, please note that if you want to receive help you should adhere to the posting guidelines. It is particularly critical that you provide a small reproducible example and your session information so package developers can track down the source of the error.

1.4 Citing derfinderHelper

We hope that derfinderHelper will be useful for your research. Please use the following information to cite the package and the overall approach. Thank you!

## Citation info
citation('derfinderHelper')
## 
## Collado-Torres L, Jaffe AE, Leek JT (2017). _derfinderHelper:
## derfinder helper package_. doi: 10.18129/B9.bioc.derfinderHelper
## (URL: http://doi.org/10.18129/B9.bioc.derfinderHelper),
## https://github.com/leekgroup/derfinderHelper - R package version
## 1.16.1, <URL: http://www.bioconductor.org/packages/derfinderHelper>.
## 
## Collado-Torres L, Nellore A, Frazee AC, Wilks C, Love MI, Langmead
## B, Irizarry RA, Leek JT, Jaffe AE (2017). "Flexible expressed region
## analysis for RNA-seq with derfinder." _Nucl. Acids Res._. doi:
## 10.1093/nar/gkw852 (URL: http://doi.org/10.1093/nar/gkw852), <URL:
## http://nar.oxfordjournals.org/content/early/2016/09/29/nar.gkw852>.
## 
## To see these entries in BibTeX format, use 'print(<citation>,
## bibtex=TRUE)', 'toBibtex(.)', or set
## 'options(citation.bibtex.max=999)'.

2 Introduction to derfinderHelper

derfinderHelper (Collado-Torres, Jaffe, and Leek, 2017) is a small package that was created to speed up the F-statistics approach implemented in the parent package derfinder. It contains a single function, fstats.apply(), which is used to calculate the F-statistics for a given data matrix, null and an alternative models.

The data is generally arranged in an matrix where the rows (\(n\)) are the genomic features of interest (gene-level summaries, exon-level summaries, or base-level data) and the columns (\(m\)) represent the samples. The other two main arguments for fstats.apply() are the null and alternative model matrices which are \(m \times p_0\) and \(m \times p\) where \(p_0\) is the number of covariates in the null model and \(p\) is the number of covariates in the alternative model. The models have to be nested and thus by definition \(p > p_0\). The end result is a vector of F-statistics with length \(n\), which is run length encoded for memory saving purposes.

Other arguments of fstats.apply() are related to flow in derfinder such as the scaling factor (scalefac) used, whether to subset the data (index), and if the data was separated into chunks and saved to disk to lower the memory load (lowMemDir).

Implementation-wise, adjustF is useful when the denominator of the F-statistic calculation is too small. Finally, method controls how will the F-statistics be calculated.

  • Matrix is the recommended option because it uses around half the memory load of regular and can be faster. Specially if the data was saved in this format previously by derfinder.
  • Rle uses the least amount of memory but gets very slow as the number of samples increases. Thus making it less than ideal in several cases.
  • regular uses base R to calculate the F-statistics and can require a large amount of memory. This is noticeable when using several cores to run fstats.apply() on different portions of the data.

The F-statistics for each feature \(i\) are calculated using the following formula:

\[ F_i = \frac{ (\text{RSS0}_i - \text{RSS1}_i)/(\text{df}_1 - \text{df}_0) }{ \text{adjustF} + (\text{RSS1}_i / (p - p_0 - \text{df_1}))} \]

3 Example

The following section walks through an example. However, in practice, you will probably not use this package directly and it will be used via derfinder.

3.1 Data

First lets create an example data set where we have information for 1000 features and 16 samples where samples 1 to 4 are from group A, 5 to 8 from group B, 9 to 12 from group C, and 13 to 16 from group D.

## Create some toy data
suppressPackageStartupMessages(library('IRanges'))
set.seed(20140923)
toyData <- DataFrame(
    'sample1' = Rle(sample(0:10, 1000, TRUE)),
    'sample2' = Rle(sample(0:10, 1000, TRUE)),
    'sample3' = Rle(sample(0:10, 1000, TRUE)),
    'sample4' = Rle(sample(0:10, 1000, TRUE)),
    'sample5' = Rle(sample(0:15, 1000, TRUE)),
    'sample6' = Rle(sample(0:15, 1000, TRUE)),
    'sample7' = Rle(sample(0:15, 1000, TRUE)),
    'sample8' = Rle(sample(0:15, 1000, TRUE)),
    'sample9' = Rle(sample(0:20, 1000, TRUE)),
    'sample10' = Rle(sample(0:20, 1000, TRUE)),
    'sample11' = Rle(sample(0:20, 1000, TRUE)),
    'sample12' = Rle(sample(0:20, 1000, TRUE)),
    'sample13' = Rle(sample(0:100, 1000, TRUE)),
    'sample14' = Rle(sample(0:100, 1000, TRUE)),
    'sample15' = Rle(sample(0:100, 1000, TRUE)),
    'sample16' = Rle(sample(0:100, 1000, TRUE))
)

## Lets say that we have 4 groups
group <- factor(rep(toupper(letters[1:4]), each = 4))

## Note that some groups have higher coverage, we can adjust for this in the model
sampleDepth <- sapply(toyData, sum)
sampleDepth
##  sample1  sample2  sample3  sample4  sample5  sample6  sample7  sample8 
##     4753     5009     4829     4969     7470     7624     7304     7380 
##  sample9 sample10 sample11 sample12 sample13 sample14 sample15 sample16 
##    10387     9644     9795     9748    49419    50509    48726    50448

3.2 Models

Next we create the model matrices for our example data set. Lets say that we want to calculate F-statistics comparing the alternative hypothesis that the group coefficients are not 0 versus the null hypothesis that they are equal to 0, when adjusting for the sample depth.

To do so, we create the nested models.

## Build the model matrices
mod <- model.matrix(~ sampleDepth + group)
mod0 <- model.matrix(~ sampleDepth)

## Explore them
mod
##    (Intercept) sampleDepth groupB groupC groupD
## 1            1        4753      0      0      0
## 2            1        5009      0      0      0
## 3            1        4829      0      0      0
## 4            1        4969      0      0      0
## 5            1        7470      1      0      0
## 6            1        7624      1      0      0
## 7            1        7304      1      0      0
## 8            1        7380      1      0      0
## 9            1       10387      0      1      0
## 10           1        9644      0      1      0
## 11           1        9795      0      1      0
## 12           1        9748      0      1      0
## 13           1       49419      0      0      1
## 14           1       50509      0      0      1
## 15           1       48726      0      0      1
## 16           1       50448      0      0      1
## attr(,"assign")
## [1] 0 1 2 2 2
## attr(,"contrasts")
## attr(,"contrasts")$group
## [1] "contr.treatment"
mod0
##    (Intercept) sampleDepth
## 1            1        4753
## 2            1        5009
## 3            1        4829
## 4            1        4969
## 5            1        7470
## 6            1        7624
## 7            1        7304
## 8            1        7380
## 9            1       10387
## 10           1        9644
## 11           1        9795
## 12           1        9748
## 13           1       49419
## 14           1       50509
## 15           1       48726
## 16           1       50448
## attr(,"assign")
## [1] 0 1

3.3 Get F-statistics

Finally, we can calculate the F-statistics using fstats.apply().

library('derfinderHelper')
fstats <- fstats.apply(data = toyData, mod = mod, mod0 = mod0, scalefac = 1)
fstats
## numeric-Rle of length 1000 with 1000 runs
##   Lengths:                  1                  1 ...                  1
##   Values :   4.10142714702069   0.37182716905003 ...  0.409243681977204

We can then proceed to use this information in derfinder or in any way you like.

4 Details

We created derfinderHelper for calculating F-statistics using SnowParam() from BiocParallel (Morgan, Obenchain, Lang, Thompson, et al., 2018). Using this form of parallelization requires loading the necessary packages in the child processes. Because derfinder takes a long time to load, we shipped off fstats.apply() to its own package to improve the speed of the calculations while retaining the memory advantages of SnowParam() over MulticoreParam().

Note that transforming the data from a DataFrame to a dgCMatrix takes some time, so the most efficient performance is achieved when the data is converted at the beginning instead of at every permutation calculation. This is done in derfinder::preprocessCoverage() when lowMemDir is specified.

5 Reproducibility

This package was made possible thanks to:

Code for creating the vignette

## Create the vignette
library('rmarkdown')
system.time(render('derfinderHelper.Rmd', 'BiocStyle::html_document'))

## Extract the R code
library('knitr')
knit('derfinderHelper.Rmd', tangle = TRUE)
## Clean up
file.remove('derfinderHelperRef.bib')
## [1] TRUE

Date the vignette was generated.

## [1] "2018-12-03 20:16:32 EST"

Wallclock time spent generating the vignette.

## Time difference of 3.722 secs

R session information.

## ─ Session info ───────────────────────────────────────────────────────────────────────────────────────────────────────
##  setting  value                                      
##  version  R version 3.5.1 Patched (2018-07-12 r74967)
##  os       Ubuntu 16.04.5 LTS                         
##  system   x86_64, linux-gnu                          
##  ui       X11                                        
##  language (EN)                                       
##  collate  C                                          
##  ctype    en_US.UTF-8                                
##  tz       America/New_York                           
##  date     2018-12-03                                 
## 
## ─ Packages ───────────────────────────────────────────────────────────────────────────────────────────────────────────
##  package         * version date       lib source        
##  assertthat        0.2.0   2017-04-11 [2] CRAN (R 3.5.1)
##  backports         1.1.2   2017-12-13 [2] CRAN (R 3.5.1)
##  bibtex            0.4.2   2017-06-30 [2] CRAN (R 3.5.1)
##  BiocGenerics    * 0.28.0  2018-12-03 [2] Bioconductor  
##  BiocManager       1.30.4  2018-11-13 [2] CRAN (R 3.5.1)
##  BiocStyle       * 2.10.0  2018-12-03 [2] Bioconductor  
##  bookdown          0.7     2018-02-18 [2] CRAN (R 3.5.1)
##  cli               1.0.1   2018-09-25 [2] CRAN (R 3.5.1)
##  crayon            1.3.4   2017-09-16 [2] CRAN (R 3.5.1)
##  derfinderHelper * 1.16.1  2018-12-03 [1] Bioconductor  
##  digest            0.6.18  2018-10-10 [2] CRAN (R 3.5.1)
##  evaluate          0.12    2018-10-09 [2] CRAN (R 3.5.1)
##  htmltools         0.3.6   2017-04-28 [2] CRAN (R 3.5.1)
##  httr              1.3.1   2017-08-20 [2] CRAN (R 3.5.1)
##  IRanges         * 2.16.0  2018-12-03 [2] Bioconductor  
##  jsonlite          1.5     2017-06-01 [2] CRAN (R 3.5.1)
##  knitcitations   * 1.0.8   2017-07-04 [2] CRAN (R 3.5.1)
##  knitr             1.20    2018-02-20 [2] CRAN (R 3.5.1)
##  lattice           0.20-38 2018-11-04 [2] CRAN (R 3.5.1)
##  lubridate         1.7.4   2018-04-11 [2] CRAN (R 3.5.1)
##  magrittr          1.5     2014-11-22 [2] CRAN (R 3.5.1)
##  Matrix            1.2-15  2018-11-01 [2] CRAN (R 3.5.1)
##  plyr              1.8.4   2016-06-08 [2] CRAN (R 3.5.1)
##  R6                2.3.0   2018-10-04 [2] CRAN (R 3.5.1)
##  Rcpp              1.0.0   2018-11-07 [2] CRAN (R 3.5.1)
##  RefManageR        1.2.0   2018-04-25 [2] CRAN (R 3.5.1)
##  rmarkdown         1.10    2018-06-11 [2] CRAN (R 3.5.1)
##  rprojroot         1.3-2   2018-01-03 [2] CRAN (R 3.5.1)
##  S4Vectors       * 0.20.1  2018-12-03 [2] Bioconductor  
##  sessioninfo     * 1.1.1   2018-11-05 [2] CRAN (R 3.5.1)
##  stringi           1.2.4   2018-07-20 [2] CRAN (R 3.5.1)
##  stringr           1.3.1   2018-05-10 [2] CRAN (R 3.5.1)
##  withr             2.1.2   2018-03-15 [2] CRAN (R 3.5.1)
##  xfun              0.4     2018-10-23 [2] CRAN (R 3.5.1)
##  xml2              1.2.0   2018-01-24 [2] CRAN (R 3.5.1)
##  yaml              2.2.0   2018-07-25 [2] CRAN (R 3.5.1)
## 
## [1] /tmp/RtmpX5455Z/Rinst20f669842808
## [2] /home/biocbuild/bbs-3.8-bioc/R/library

6 Bibliography

This vignette was generated using BiocStyle (Oleś, Morgan, and Huber, 2018) with knitr (Xie, 2014) and rmarkdown (Allaire, Xie, McPherson, Luraschi, et al., 2018) running behind the scenes.

Citations made with knitcitations (Boettiger, 2017).

[1] J. Allaire, Y. Xie, J. McPherson, J. Luraschi, et al. rmarkdown: Dynamic Documents for R. R package version 1.10. 2018. URL: https://CRAN.R-project.org/package=rmarkdown.

[1] D. Bates and M. Maechler. Matrix: Sparse and Dense Matrix Classes and Methods. R package version 1.2-15. 2018. URL: https://CRAN.R-project.org/package=Matrix.

[1] C. Boettiger. knitcitations: Citations for ‘Knitr’ Markdown Files. R package version 1.0.8. 2017. URL: https://CRAN.R-project.org/package=knitcitations.

[1] L. Collado-Torres, A. E. Jaffe and J. T. Leek. derfinderHelper: derfinder helper package. https://github.com/leekgroup/derfinderHelper - R package version 1.16.1. 2017. DOI: 10.18129/B9.bioc.derfinderHelper. URL: http://www.bioconductor.org/packages/derfinderHelper.

[1] G. Csárdi, R. core, H. Wickham, W. Chang, et al. sessioninfo: R Session Information. R package version 1.1.1. 2018. URL: https://CRAN.R-project.org/package=sessioninfo.

[1] M. Morgan, V. Obenchain, M. Lang, R. Thompson, et al. BiocParallel: Bioconductor facilities for parallel evaluation. R package version 1.16.2. 2018. URL: https://github.com/Bioconductor/BiocParallel.

[1] A. Oleś, M. Morgan and W. Huber. BiocStyle: Standard styles for vignettes and other Bioconductor documents. R package version 2.10.0. 2018. URL: https://github.com/Bioconductor/BiocStyle.

[1] H. Pagès, M. Lawrence and P. Aboyoun. S4Vectors: S4 implementation of vector-like and list-like objects. 2017. DOI: 10.18129/B9.bioc.S4Vectors.

[1] R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. Vienna, Austria, 2018. URL: https://www.R-project.org/.

[1] H. Wickham. “testthat: Get Started with Testing”. In: The R Journal 3 (2011), pp. 5–10. URL: https://journal.r-project.org/archive/2011-1/RJournal_2011-1_Wickham.pdf.

[1] Y. Xie. “knitr: A Comprehensive Tool for Reproducible Research in R”. In: Implementing Reproducible Computational Research. Ed. by V. Stodden, F. Leisch and R. D. Peng. ISBN 978-1466561595. Chapman and Hall/CRC, 2014. URL: http://www.crcpress.com/product/isbn/9781466561595.