Fig. 4: Impact of sample size on estimated effect size.

A Percentage of DEGs in each trial whose effect size (2^abs(log2FC), perturbation in either direction) in the sub-sampled trial exceeds the effect size in the 30-vs-30 comparison Sample size N is shown along the x-axis. The dotted line at 50% indicates no bias, where a gene is equally likely to over- or under-estimate the effect. B Effect size estimates for a representative gene (Trex1) in liver. For each N (x-axis), the Het-vs-WT log₂ fold change estimate of Trex1 is shown separately for statistically significant (cyan), or not significant (red) trials. Statistical significance was defined as having amultiple hypothesis adjusted P-value of 0.05 or lower, as calculated using the DeSeq2 package’s negative binomial model. With small N, fold change estimates from significant trials overestimate that of the gold standard (solid horizontal line), but approach it as N increases. Non-significant trials by contrast show no bias at small N.