Fig. 2

High variability in gene expression levels compromises performance. a Our method performs less well on cell-line G&T-sequencing data than on the mouse embryos. All cells were considered for the 8-cell embryos and HCC38-BL data. Trisomy 21 cells were downsampled to a ratio of 1 T21 cell : 4 control cells (normal ploidy chromosome 21), to ensure that these aneuploidies were in the minority and therefore can be detected. b The datasets with poor performance show more variable gene expression profiles. For 500 genes selected at random from each dataset (navy: HCC38-BL cell line; yellow: Reversine-treated 8-cell embryos; cyan: trisomy 21 iPS derived neurons) we plot the (log) standard deviation of expression (y−axis) against the (log) mean expression (x−axis). A linear model was fitted separately for each dataset using genes with a median count (per million reads) of at least 50 and overlaid. c Simulated datasets with different dispersion parameters are shown. We simulated four datasets to assess the impact of gene expression variability on the performance of our method. Genes from each simulation are shown and the different dispersion parameters used in the simulation are noted. The regression lines from the fit in 2b are overlaid. d As the data become more variable, the performance of our method degrades. For simulations with variability comparable to the HCC38-BL and Trisomy 21 neuron datasets, the sensitivity and precision are considerably impacted. Reported values are the mean of 10 simulations