skip to navigation
skip to content

LIMBR 0.2.7

Learning and Imputation for Mass-spec Bias Reduction

LIMBR provides a streamlined tool set for imputation of missing data followed by modelling and removal of batch effects. The software was designed for proteomics datasets, with an emphasis on circadian proteomics data, but can be applied to any time course or blocked experiments which produce large amounts of data, such as RNAseq. The two main classes are imputable, which performs missing data imputation, and sva, which performs modelling and removal of batch effects.


Decreasing costs and increasing ambition are resulting in larger Mass-spec (MS) experiments. MS experiments have a few limitations which are exacerbated by this increasing scale, namely batch effects and missing data. Many downstream statistical analyses require complete cases for analysis, however, MS produces some missing data at random meaning that as the number of experiments increase the number of peptides rejected due to missing data actually increases. This is obviously not good, but fortunately there is a solution! If the missing data for observations missing only a small number of data points are imputed this issue can be overcome and that’s the first thing that LIMBR does. The second issue with large scaler MS experiments is batch effects. As the number of samples increases, the number of batches necessary for sample processing also increases. Batch effects from sample processing are known to have a large effect on MS data and increasing the number of batches means more batch effects and a higher proportion of observations affected by at least one batch effect. Here LIMBR capitolizes on the larger amount of data and the known correlation structure of the data set to model these batch effects so that they can be removed.


  • KNN based imputation of missing data.
  • SVD based modelling and removal of batch effects.
  • Built for circadian and non-circadian time series as well as block designs

Example Usage


pip install limbr

How to Use?


  • Switch to long format files for greater interoperability and more easily specified file format.
  • Add unit tests to docstrings where possible.


K nearest neighbors as an imputation method was originally proposed by Gustavo Batista in 2002 ( and has seen a great deal of success since.

The sva based methods build on work for micro-array datasets by Jeffrey Leek, with particular reliance on his PhD Thesis from the University of Washington (


© 2017 Alexander M. Crowell: BSD-3

File Type Py Version Uploaded on Size
LIMBR-0.2.7.tar.gz (md5) Source 2018-01-18 11KB