We first selected 1000 informative genes by fitting a support-vector regression to the coefficient of variation (CV) as a function of the mean, and selecting genes having the greatest offset from the fitted curve; this would correspond to genes with higher-than-expected variance. We excluded the sex-specific genes Xist and Tsix. We normalized each cell to a sum of 5,000 molecules (UMIs), then log-transformed and subtracted the mean (per gene).
