Covariate balanced allocation of samples to batches to mitigate the impacts of technical variability

Abstract Motivation Performing assays on large numbers of samples requires their analysis in distinct batches, which commonly affects the measurements made in a systematic way. Analytical approaches can correct for such batch effects, however for this to be possible the batch should not be confounded with either the independent variable or any covariates relevant to the analysis. Thus, how samples are distributed across batches influences subsequent analytic conclusions. Results We present SampleAllocateR, a tool that uses optimizsation methods from machine learning to optimally allocate a preselected set of samples to experimental batches in a way that statistically balances specified covariates between batches. This will result in better statistical estimates of the effect of not only the technical batch effects but also the other specified covariates upon the results of the experiment. SampleAllocateR accepts both continuous and categorical covariates, supports blocking of related samples (such as those from the same subject), and is tolerant of missing metadata. Availability and implementation SampleAllocateR is freely available as an open source R package at https://github.com/CardiacProteomics/SampleAllocateR.