Fig. 2

Schematic overview of CYANO-MLP phyloclassification workflow. a The cyanobacterial phylogeny from [10] was used to define cyanobacterial clades for which we estimated tRNA CIFs. Cyanobacterial clades, indicated by background colors, were named according to [10], except for clade B2+3, which combines clades B2 and B3. Clades with grey backgrounds were excluded from analysis because of limited genome sample sizes. Next, tRNA genes are predicted from cyanobacterial genomes and combined by clade for function logo estimation [31]. b Example function logos, for Uracil, with background colors corresponding to each cyanobacterial clade. c For a genome g to be classified, tRNA gene complements T_g are predicted and scored against the logos for each clade, to produce input score vectors for CYANO-MLP training and classifications. d The architecture of the artificial neural network CYANO-MLP, and a classification probability vector output from CYANO-MLP represented as a stacked bar chart