Genelists were derived from the OVCAR data using the top 250 genes (by size of fold change) significantly regulated for either RIα or RIIβ cells versus parental cells as previously described [ 30] and these genelists were tested against all calculated changes in gene expression between day 2 9 and day 0 1 from the myotube dataset for concordance by Gene Set Matrix Analysis (GSMA) analysis [ 19].
Alterations were considered significant if it passed the following filtering criteria: (i) q < 0.25 (individual dataset), (ii) q < 0.05 (merged dataset), and (iii) concordance in 2 or more datasets.
First, all the WGS datasets achieved high concordance rates of the genotypes with those genotyped by SNP microarrays31,32 (≥99.97%), which provides confidence in calling of common variants.
These datasets provide significant concordance with each other, with previous GWA datasets for substance dependence, and with dbGAP data that has been available only after the results of the current study were analyzed.
Using these datasets, an overall concordance between level of genetic differentiation and linkage mapping signal was evaluated.
F2 samples typed in both the HincII and RsaI datasets had a concordance of 97.89% ±1.00% (SD) on a genomewide, nucleotide level.
While there are positive signs, such as that in independent datasets there is concordance amongst different gene expression signatures, it is not clear whether gene expression profiling simply captures already well known biological characteristics derived from careful phenotypic observations [ 2].
In the lung adenocarcinoma dataset, we quantified concordance in two different ways.
Looking at differences between 28- and 32-joint counts in our prospective observational dataset, we found concordance rates as well as sensitivity of 28-joint counts in the order of 95% and above using the 32-joint count as a gold standard.
Where a gene was represented several times on a given platform, an appropriate rule was used to select the "best" gene probe in a manner not biased towards detecting patterns of concordance between datasets (for human lung tumor datasets, the probe with the greatest variation; for MCF-7 datasets, the probe with the greatest difference in either direction by t test between EGFR and control).
