Improved detection of clinically relevant fusion transcripts in cancer by machine learning classification

BMC Genomics

Table 4 Performance metrics for machine learning-based filtering compared to classical filtering methods on fusions detected with Arriba in an independent fusion dataset of 249 TNBC samples from the SCAN-B cohort

Metric	Classifier	High confidence	High+med confidence	RNA disc. mates > 2	High + med confidence and RNA disc. mates > 2	In-frame only
precision	0.74	0.70	0.68	0.63	0.72	0.69
recall	0.70	0.54	0.72	0.69	0.54	0.20
specificity	0.67	0.68	0.55	0.46	0.72	0.88
accuracy	0.69	0.60	0.64	0.59	0.62	0.49
f1 score	0.72	0.61	0.70	0.66	0.62	0.31
kappa	0.37	0.22	0.27	0.15	0.25	0.07
log loss	0.83	-	-	-	-	-
roc auc	0.76	-	-	-	-	-
pr auc	0.79	-	-	-	-	-
brier score	0.26	-	-	-	-	-

ISSN: 1471-2164