Motivation: Reverse phase protein arrays (RPPA) measure the relative expression levels of a protein in many samples simultaneously. A set of identically spotted arrays can be used to measure the levels of more than one protein. Protein expression within each sample on an array is estimated by borrowing strength across all the samples, but using only within array information. When comparing across slides, it is essential to account for sample loading, the total amount of protein printed per sample. Currently, total protein is estimated using either a housekeeping protein or the sample median across all slides. When the variability in sample loading is large, these methods are suboptimal because they do not account for the fact that the protein expression for each slide is estimated separately.
Results: We propose a new normalization method for RPPA data, called variable slope (VS) normalization, that takes into account that quantification of RPPA slides is performed separately. This method is better able to remove loading bias and recover true correlation structures between proteins.
Availability: Code to implement the method in the statistical package R and anonymized data are available at (http://bioinformatics.mdanderson.org/supplements.html).