Defining a landscape of molecular phenotypes using a simple single sample scoring method

Abstract

BackgroundGene set scoring provides a useful approach for quantifying concordance between sample transcriptomes and selected molecular signatures. Most methods use information from all samples to score an individual sample, leading to unstable scores in small data sets and introducing biases from sample composition across a data set (e.g. varying numbers of samples for different cancer subtypes). To address these issues we have developed a truly single sample scoring method, and associated R/Bioconductor package singscore.\n\nResultsWe have developed a rank-based single sample scoring method, implemented as a Bioconductor package. We use multiple cancer data sets to compare it against widely-used scoring methods, including GSVA, z-scores, PLAGE, and ssGSEA. Our approach does not depend upon background samples and thus the scores are stable regardless of the composition and number of samples in the gene expression data set. In contrast, scores obtained by GSVA, z-score, PLAGE and ssGSEA can be unstable when less data are available (nsamples < 25). We show that the computational time for singscore is faster than current implementations of GSVA and ssGSEA, and is comparable with that of z-score and PLAGE. The singscore package also produces visualisations and interactive plots that enable exploration of molecular phenotypes.\n\nConclusionsThe single sample scoring method described here is independent of sample composition in gene expression data and thus it provides stable scores that are less likely to be influenced by unwanted variation across samples. These scores can be used for dimensional reduction of transcriptomic data and the phenotypic landscapes obtained by scoring samples against multiple molecular signatures may provide insights for sample stratification.