Correcting for experiment-specific variability in expression compendia can remove underlying signals

Abstract

MotivationIn the last two decades, scientists working in different labs have assayed gene expression from millions of samples. These experiments can be combined into compendia and analyzed collectively to extract novel biological patterns. Technical variability, sometimes referred to as batch effects, may result from combining samples collected and processed at different times and in different settings. Such variability may distort our ability to interpret and extract true underlying biological patterns. As more multi-experiment, integrative analysis methods are developed and available data collections increase in size, it is crucial to determine how technical variability affect our ability to detect desired patterns when many experiments are combined ObjectiveWe sought to determine the extent to which an underlying signal was masked by technical variability by simulating compendia comprised of data aggregated across multiple experiments. MethodWe developed a generative multi-layer neural network to simulate compendia of gene expression experiments from large-scale microbial and human datasets. We compared simulated compendia before and after introducing varying numbers of sources of undesired variability. ResultsWe found that the signal from a baseline compendium was obscured when the number of added sources of variability was small. Perhaps as expected, applying statistical correction methods rescued the underlying signal in these cases. As the number of sources of variability increased, surprisingly, we observed that detecting the original signal became increasingly easier even without correction. In fact, applying statistical correction methods reduced our power to detect the underlying signal. ConclusionWhen combining a modest number of experiments, it is best to correct for experiment-specific noise. However, when many experiments are combined, statistical correction reduces ones ability to extract underlying patterns.