Fig. 2

The residual variance statistic captures variation in gene families after accounting for between-study variation. The left-hand panels (“original abundances") show filled circles representing log-RPKG abundances for gene families from the KEGG Orthology (KO), with per-study means shown in solid horizontal lines and the distance from these means shown as dashed vertical lines. The right-hand panels (“residuals") show the same gene families after fitting a linear model that accounts for these per-study means, with an accompanying density plot showing the distribution of these residuals. \(V_{g}^{\epsilon }\) values in bold underneath density plots are the calculated variances of these residuals. These gene families are sets of orthologs corresponding to the genes a tatA, b devR, c waaW, d thrC, e gspA, f tssB, g dctS, and h ecnB. Panels a,b show two invariable gene families with relatively high (a) and low (b) average abundance; similarly, panels c, d show two variable gene families with relatively low (c) and high (d) average abundances. Panels e, f show two gene families involved in secretion with similar abundances, but low (e) vs. high (f) variability. Finally, panels g, h show that both invariable (g) and variable (h) gene families can have substantial study-specific effects. (All gene families displayed were significantly (in)variable using CCoDA, FDR ≤5%)