ComBat is an R package for removing batch effects from sequencing data. The most prominent implementation can be found in the R package SVA. This FASTGenomics app uses the faster python implementation of the SVA code provided by Brent Pedersen on his GitHub page.
In the current state, the app takes no parameters. No additional covariate data is used.
The combat function is called on the input data set specified in
config/input_file_mapping.json, assumes a single batch column as specified in enum Columns.BATCHES in main.py.
The app stores the transformed matrix in an output file as specified in manifest.json. It also provides a very simple statistic for the effect of batch correction, the average distance between all batch cluster centers, included in the summary output. For an array all_batch_reps holding the mean vectors of each batch, this statistic is computed as np.sum(pdist(all_batch_reps)).
Finally, for code debugging and visual inspection of the correction results, a simple PCA embedding before and after correction is plotted.
In the currect state, this app requires pandas version 0.19.2, as stated in requirements.txt. Batch correction will crash for higher versions of pandas.
Pedersen, Brent: GitHub page for python / numpy / pandas / patsy version of ComBat for removing batch effects. https://github.com/brentp/combat.py Accessed Oct. 2017.
Bioconductor site for the original SVA package. http://bioconductor.org/packages/release/bioc/html/sva.html. Accessed Oct. 2017.
Johnson WE, Rabinovic A, Li C (2007). Adjusting batch effects in microarray expression data using Empirical Bayes methods. Biostatistics 8:118-127.
Jeffrey T. Leek, W. Evan Johnson, Hilary S. Parker, Andrew E. Jaffe and John D. Storey (). sva: Surrogate Variable Analysis. R package version 3.4.0.