Fig. 3

Training results of CYANO-MLP. a Null distribution of classification accuracy generated using 100,000 permutations of score vectors with clade labels swapped. The black line indicates expected accuracy if genomes were randomly assigned to clades and the green line indicates accuracy of CYANO-MLP using leave one out cross-validation. Normalized confusion matrices with the number of genomes used for training indicated on the right side for b CYANO-MLP and a c clade balanced version of CYANO-MLP produced by oversampling under-represented clades