Fig. 1

Ψ prediction model training using model organisms and microbiome rRNA Ψ modification. a Overview of the experiments to generate the Ψ prediction model by nanopore sequencing. b Features of a region in human 18S rRNA from Illumina sequencing and nanopore sequencing. c Features of a region in a microbial rRNA from Illumina sequencing and nanopore sequencing. d Box and Whisker plots with 1.5 times interquartile range of the 12 feature candidates of U and Ψ sites derived from nanopore sequencing. Ins, insertion rate after the base. Ins_len, insertion length mean. Del, deletion rate after the base. Del_len, deletion length mean. Del_site, deleted site ratio (the site is in a deletion). Mis, overall mismatching ratio. Mis_A, mutation to A ratio. Mis_C, mutation to C ratio. Mis_G, mutation to G ratio. Base_qual_mean, average base quality score. Base_qual_STD, base quality score standard deviation. Base_qual_count_0, ratio of bases with a quality score 0 at a site. e Mutation preference for the Ψ sites in all rRNAs in a ternary plot. Red, Ψ sites in model organisms. Blue, Ψ sites in the microbiome. f Correlation matrix of modification state (Ψ=1, U=0) and the 12 feature candidates. The value of correlation coefficient is indicated in each box. Same labels as panel d. Label type, modification state. g ROC (receiver operating characteristic) curves of EXT models with different numbers of features included. The number of features and AUC (aera under curve) values of each model are indicated by the legend. The features are added to the model in the order of their correlation with the modification state indicated in panel f. For example, 1 feature means “mis_C”, 2 features means “mis_C” and ”mis”, and so on. h ROC curve of the testing set predicted by the optimized EXT model. The AUC value is indicated in the graph