Fig. 3

Functional profiles characterization. a KO and GH-related AFT frequencies are first gathered to show the distribution of KO and GH in each profile (top central pie chart). Then, the frequency of each AFT is renormalized by KO or GH/PL total frequency, and displayed in pie-charts for KO (left) and GH/PL (right) after clustering by functional modules (color coded; see Fig. 1a for the functional modules). The number of the KO or GH-related AFT is displayed in its corresponding pie-chart sector (radially, inner zone) when its frequency is higher than 3% in the profile. b Average profile contribution in AFT counts. Average profile contribution for AFT j and profile i is computed as the proportion of average AFT counts provided by the profile i with \(\bar{W}^{(AFT)}_{train,i} H^{(AFT)}_{ij} / \bar{X}^{(AFT)}_{train,j}\), where \(\bar{W}^{(AFT)}_{train}\) and \(\bar{X}^{(AFT)}_{train}\) are introduced in the “Inference of functional profiles” section. Finally, contributions are stacked by AFT in bar plots and ordered by functional modules. The residual \(1-\sum _{i=1}^4\bar{W}^{(AFT)}_{train,i} H^{(AFT)}_{ij} / \bar{X}^{(AFT)}_{train,j}\) is plotted in gray. Dotted gray lines indicate the value of \(\bar{X}^{(AFT)}_{train,j}\) measuring the average AFT frequency (y log-scale on the right)