Multivariate models from RNA-Seq SNVs yield candidate molecular targets for biomarker discovery: SNV-DA

BMC Genomics

Table 3 NSCLC model performances

Model	Tested Range of K, Every Nth	Opt. K	AUC [95 % CI]	P-value	Pred. Accuracy	DF Sens.	R Sens.
nonsynonymous exonic	10–1000, 40	841	0.874 [0.845–0.903]	0.016	0.803	0.836	0.763
3’ UTR	10–1000, 40	46	0.720 [0.650–0.789]	>0.10	0.626	0.626	0.626
up/downstream	10–400, 15	138	0.708 [0.690 – 0.725]	>0.10	0.653	0.690	0.615
all-SNVs	10–1000, 40	139	0.643 [0.615–0.671]	>0.10	0.587	0.679	0.474
intronic	10–1000, 40	59	0.634 [0.599–0.669]	>0.10	0.607	0.558	0.667
RNA-editing	10–400, 10	36	0.615 [0.577 – 0.653]	>0.10	0.573	0.503	0.659
exonic	10-1000, 40	54	0.580 [0.552–0.609]	>0.10	0.450	0.612	0.252
intergenic	10–1000, 40	60	0.561 [0.519–0.602]	>0.10	0.487	0.642	0.296
ncRNA	10–1000, 40	36	0.556 [0.520–0.591]	>0.10	0.547	0.721	0.333
5’ UTR	10–750, 30	16	0.242 [0.215–0.268]	>0.10	0.290	0.394	0.163
gene expression	10–1000, 40	592	0.824 [0.803–0.845]	0.068	0.740	0.764	0.711

The range of K tested, the optimal value of K, AUC and 95 % confidence interval, p value from 1000 iteration permutation tests, predictive accuracy, and classification sensitivities of the top-performing models by genomic region

ISSN: 1471-2164