Abstract
The solution to identify a reliable molecular signature in transcriptomics high-throughput experiments is to increase sample size by combining independent but related studies. However, those data sets are generated using different protocols and technological platforms, which results in unwanted systematic variation that strongly confounds the integrative analysis results. We introduce a Multi-variate INTegrative method, MINT, that identifies a highly reproducible, accurate and predictive gene signature to classify sample phenotypes while accounting for platform and study variation. MINT led to superior and unbiased classification performance compared to other existing methods, and identified highly relevant gene signatures when integrating two multi-transcriptomics studies.
