Removing Unwanted Variation from High Dimensional Data with Negative Controls

High dimensional data suffer from unwanted variation, such as the batch effects common in microarray data. Unwanted variation complicates the analysis of high dimensional data, leading to high rates of false discoveries, high rates of missed discoveries, or both. In many cases the factors causing the unwanted variation are unknown and must be inferred from the data. In such cases, negative controls may be used to identify the unwanted variation and separate it from the wanted variation. We present a new method, RUV-4, to adjust for unwanted variation in high dimensional data with negative controls. RUV-4 may be used when the goal of the analysis is to determine which of the features are truly associated with a given factor of interest. One nice property of RUV-4 is that it is relatively insensitive to the number of unwanted factors included in the model; this makes estimating the number of factors less critical. We also present a novel method for estimating the features’ variances that may be used even when a large number of unwanted factors are included in the model and the design matrix is full rank. We name this the “inverse method for estimating variances.” By combining RUV-4 with the inverse method, it is no longer necessary to estimate the number of unwanted factors at all. Using both real and simulated data we compare the performance of RUV-4 with that of other adjustment methods such as SVA, LEAPP, ICE, and RUV-2. We find that RUV-4 and its variants perform as well or better than other methods.