An investigation of biomarkers derived from legacy microarray data for their utility in the RNA-seq era

Su, Zhenqiang; Fang, Hong; Hong, Huixiao; Shi, Leming; Zhang, Wenqian; Zhang, Wenwei; Zhang, Yanyan; Dong, Zirui; Lancashire, Lee J; Bessarabova, Marina; Yang, Xi; Ning, Baitang; Gong, Binsheng; Meehan, Joe; Xu, Joshua; Ge, Weigong; Perkins, Roger; Fischer, Matthias; Tong, Weida

doi:10.1186/s13059-014-0523-y

Table 5 Performance metrics for the assessment of cross-platform transferability of signature genes of k-NN models based on the SEQC NB data

From: An investigation of biomarkers derived from legacy microarray data for their utility in the RNA-seq era

Direction	Endpoint	Gene set	Microarray models Is predict microarray validation samples				RNA-Seq models predict RNA-Seq validation samples				T-index
			Accuracy		AUC		Accuracy		AUC
			Mean	95% CI	Mean	95% CI	Mean	95% CI	Mean	95% CI
From microarray to RNA-Seq	A*	A	0.732	0.667-0.775	0.708	0.645-0.758	0.728	0.653-0.771	0.696	0.627-0.746	0.729
		B	0.721	0.655-0.759	0.691	0.631-0.733	0.714	0.647-0.759	0.678	0.614-0.726	0.716
		C	0.696	0.639-0.735	0.664	0.609-0.704	0.702	0.654-0.747	0.670	0.631-0.713	0.700
	B*	A	0.780	0.747-0.811	0.667	0.615-0.72	0.787	0.755-0.815	0.648	0.601-0.694	0.786
		B	0.777	0.743-0.807	0.643	0.585-0.704	0.794	0.755-0.827	0.659	0.580-0.725	0.791
		C	0.789	0.739-0.827	0.678	0.613-0.732	0.788	0.747-0.819	0.660	0.620-0.711	0.788
	C*	A	0.978	0.971-0.992	0.978	0.971-0.992	0.992	0.992-0.992	0.992	0.992-0.992	0.992
		B	0.989	0.983-0.992	0.989	0.984-0.992	0.989	0.983-0.992	0.989	0.984-0.992	0.989
		C	0.931	0.900-0.954	0.935	0.905-0.958	0.988	0.975-0.992	0.988	0.976-0.992	0.984
	D*	A	0.934	0.882-0.971	0.920	0.863-0.962	0.921	0.882-0956	0.895	0.848-0.940	0.921
		B	0.947	0.824-0.978	0.938	0.798-0.973	0.933	0.772-0.978	0.915	0.737-0.967	0.934
		C	0.915	0.838-0.956	0.911	0.841-0.961	0.921	0.853-0.963	0.914	0.840-0.956	0.920
	E*	A	0.624	0.533-0.700	0.534	0.463-0.615	0.606	0.522-0.689	0.537	0.460-0.617	0.612
		B	0.562	0.478-0.623	0.507	0.417-0.588	0.569	0.500-0.633	0.519	0.443-0.595	0.566
		C	0.607	0.511-0.689	0.513	0.437-0.603	0.599	0.511-0.689	0.515	0.438-0.603	0.602
	F*	A	0.513	0.444-0.589	0.513	0.446-0.585	0.510	0.433-0.589	0.510	0.426-0.589	0.511
		B	0.507	0.456-0.567	0.511	0.452-0.572	0.490	0.422-0.556	0.498	0.431-0.562	0.498
		C	0.532	0.456-0.611	0.534	0.460-0.609	0.527	0.444-0.611	0.531	0.448-0.611	0.529
From RNA-Seq to microarray	A*	A	0.701	0.614-0.759	0.669	0.590-0.732	0.709	0.643-0.759	0.671	0.603-0.727	0.703
		B	0.680	0.584-0.741	0.644	0.543-0.708	0.693	0.604-0.747	0.651	0.568-0.710	0.684
		C	0.719	0.647-0.767	0.691	0.618-0.748	0.730	0.648-0.771	0.698	0.621-0.743	0.722
	B*	A	0.775	0.715-0.819	0.662	0.576-0.737	0.775	0.733-0.811	0.640	0.579-0.696	0.775
		B	0.777	0.715-0.819	0.639	0.552-0.713	0.785	0.735-0.825	0.640	0.566-0.718	0.778
		C	0.792	0.749-0.823	0.681	0.621-0.757	0.790	0.753-0.819	0.659	0.606-0.728	0.791
	C*	A	0.971	0.967-0.992	0.972	0.969-0.992	0.984	0.983-0.992	0.985	0.984-0.992	0.971
		B	0.939	0.900-0.950	0.943	0.901-0.954	0.990	0.988-0.992	0.990	0.988-0.992	0.939
		C	0.987	0.975-0.992	0.987	0.975-0.992	0.987	0.971-0.992	0.987	0.972-0.992	0.987
	D*	A	0.914	0.809-0.971	0.899	0.792-0.957	0.927	0.868-0.963	0.910	0.841-0.951	0.915
		B	0.918	0.838-0.963	0.907	0.825-0.962	0.928	0.860-0.971	0.908	0.836-0.957	0.918
		C	0.937	0.875-0.978	0.923	0.853-0.973	0.933	0.882-0.971	0.913	0.882-0.962	0.937
	E*	A	0.598	0.456-0.689	0.506	0.414-0.606	0.598	0.444-0.700	0.522	0.400-0.630	0.598
		B	0612	0.511-0.700	0.513	0.432-0.602	0.588	0.489-0.678	0.510	0.415-0.613	0.603
		C	0.582	0.478-0.667	0.512	0.412-0.633	0.596	0.500-0.689	0.522	0.432-0.615	0.587
	F*	A	0.507	0.433-0.600	0.502	0.421-0.590	0.498	0.400-0.600	0.496	0.406-0.597	0.503
		B	0.500	0.411-0.595	0.503	0.414-0.596	0.493	0.400-0.589	0.503	0.411-0.601	0.496
		C	0.523	0.456-0.600	0.527	0.452-0.602	0.498	0.405-0.578	0.505	0.411-0.59	0.511

A*: A_EFS_All; B*: B_OS_All; C*: C_SEX_All; D*: D_FAV_All; E*: E_EFS_HR; F*: F_OS_HR; AUC: Area under ROC curve; CI: Confidence interval; 95% CI was calculated from the bootstrap estimation. The upper-right and lower-left regions are for the untrained models built using cross-platform transferred signature genes, while the upper-left and lower-right regions are for the models originally trained.

Back to article page

ISSN: 1474-760X

Contact us

Submission enquiries: editorial@genomebiology.com
General enquiries: info@biomedcentral.com

Genome Biology

Contact us