Fig. 2

Workflow of the standardized individual dataset analysis. The analysis of the nine datasets was performed in a consistent manner for each dataset using ad hoc R scripts. It included the first step of data preparation followed by a second step of data analysis. GEPs were processed using standard quality control tools to obtain normalized, probeset-level expression data. For raw datasets derived from affymetrix chips, Robust Multi-Array Average expression measure (rma) was used as normalization method. All analyses were conducted at probeset level. Probesets were annotated to gene symbol and gene ENTREZ using chip-specific annotations. Quality controls were performed on raw data using RLE and NUSE plots. The distribution of the QSC and ASC samples according to their GEPs was explored using hierarchical clustering of the Euclidean distance and principal component analysis (Additional file 1: Figure S1). Statistically, differentially expressed genes (DEGs) were identified between the ASC and the QSC groups using the linear model implemented by the Limma R package [10]. Gene set enrichment analysis was based on three gene set collections from the mouse version of the Molecular Signatures Database MSigDB v6.0 [12, 13]: (1) Hallmark, which summarizes and represents specific well-defined biological states or processes displaying a coordinate gene expression; (2) KEGG canonical pathways, derived from the Kyoto Encyclopedia of Genes and Genomes [14]; and (3) Reactome canonical pathways from the curated and peer-reviewed pathway database [15]. To test for the enrichment of these gene sets, the competitive gene set test CAMERA [16] was used