SQANTI filter performance on mouse data set. (A) Representative examples of RT-PCR validation experiments. Three PCR conditions were assessed: oligo(dT) template at 42°C and random hexamers (RH) template at 42°C and at 50°C. (A1) Example of a FSM transcript with a noncanonical SJ successfully amplified at each PCR condition. (A2) Example of a NNC transcript with a noncanonical SJ that failed to be amplified in the oligo(dT) condition. (A3) Example of NNC transcript with noncanonical SJ amplified at oligo(dT) but not at RH conditions. (B) ROC curves of the SQANTI ML filter for the test set (blue line) and for the set of novel isoforms assayed by RT-PCRs (red line). (C) Summary of the performances of the SQANTI filter, the noncanonical filter, and the SpliceGrapher filter for the set of novel isoforms assayed by RT-PCR. (nc filter) Noncanonical filter, (TP) True Positive, (TN) True Negative, (FP) False Positive, (FN) False Negative, (FDR) False Discovery Rate. (D) Comparison of quality features in the FSM and NNC categories after the SQANTI, nc, and SpliceGrapher filters. Statistical differences by Fisher's exact test (FET), (***) P < 0.001, (ns) not significant. (E) Composition of SQANTI transcript categories in the mouse before and after the SQANTI filter.
