Fig. 4

Legionella ASVs group according to sodium concentrations. Next-generation sequencing (NGS) was performed using Legionella-specific 16S primers. Raw data were pre-filtered to retain Legionella amplicon sequence variants (ASVs), present at a minimum of 20 or 2000 reads across all samples, in a minimum of 2 samples. Water (n = 40) and biofilm (n = 25) samples were analyzed separately. Spring clusters: Tiberias Hot Springs (THS, n = 20), Haon borehole (n = 4), Fuliya (n = 8) and Tabgha (n = 33). ASVs are represented by the last 5 letter-number code (for full-length codes and corresponding representative sequence, see Supplementary Data 1. Heatmap presenting abundant Legionella ASVs (2000 reads), in a water and b biofilm samples. ASVs were grouped according to Hierarchical clustering and ordered according to sodium concentrations. The heatmap color scale represents high (green to yellow) and low (blue to purple) centered-scaled values of Legionella ASVs relative abundance. Sodium concentrations and spring clusters are shown to the left of the heatmap. Additionally, ASVs associated with high sodium concentrations were also predicted by a random forest procedure (see “Random forest analysis” section). ASVs that were identified in the heatmap as well as the random forest procedure are colored red. Legionella ASVs associated with high sodium concentrations in the random forest are presented for c water and d biofilm samples. ASV importance was predicted by the machine learning algorithm. Scaled-centered relative abundances (20 reads) were used to evaluate ASV importance, based on the decrease in mean accuracy (%IncMSE)