Extended Data Fig. 5: Optimization of PORE-cupine using 11 RNAs as training set.
From: Determination of isoform-specific RNA structure with nanopore long reads
a, Scatterplot showing the distribution of normalized base reactivity between N = 2 biological replicates of modified Tetrahymena RNA. R = 0.97, CI95% = [0.97,0.98] (Pearson correlation). P-value=2.5×10-262, two-tailed Student’s T-test. b, Distribution of current mean and standard deviation for a unimodal (left) and bimodal (right) position in two biological replicates. c, AUC-ROC performance of the correlation of NAI-N3 reactivities of the training set based on PORE-cupine versus footprinting from 11 transcripts. d,e, Comparison of PORE-cupine reactivity and traditional footprinting. Two replicates of gels were shown for Tetrahymena RNA (d, R = 0.80) and lysine riboswitch (e, R = 0.74). Lane 1 of the footprinting gels show A (left, Tetrahymena) or G (right, Tetrahymena and lysine) ladder. Lane 2 shows unmodified RNA, and lane 3 shows NAI-N3 modified RNA. Quantification of the bands on the gels was done using SAFA. Pearson correlation was used to compare between SAFA and PORE-cupine signals. f, List of RNAs used for training and test. g, Scatter plot of per-base reactivity in two biological replicates of the three test RNAs. P-value = 0 using two-tailed Student’s T-test. R = 0.877, CI95% = [0.87, 0.89], by Pearson correlation. h-j, Line plots showing the per-base reactivity along the length of three test RNAs, for two biological replicates. R > = 0.89, using Pearson correlation. k, Boxplot showing the performance of the SVM parameters on the 3 test RNAs, based on training on the Tetrahymena RNA (left) or on 11 RNAs (right, footprinting gels). l, AUC-ROC performance of SVM parameters on 3 test RNAs (red, based on our current 11 training RNAs) versus test RNAs after random selection of 11/14 RNAs as training, for 20 times. m, Boxplot showing the performance of all, unimodal and bimodal positions on test RNAs using AUC-ROC based on footprinting gels from 3 transcripts. In c, k-m, the middle, lower and upper boundary lines in the boxplot correspond to median, first and third quartiles. The upper whisker extends to the largest value no further than 1.5 × IQR from the hinge (where IQR is the inter-quartile range) and the lower whisker extends to the smallest value at most 1.5 × IQR of the hinge. Outliers are shown as dots.
