Table 1.

Application of multivariate outlier detection to negative and positive controls derived from MAQC and Affymetrix spike-in series, the latter with digital contamination

Negative controls
Source	No. of chips	No. of chips flagged
		PMVO-raw	PVMO-PC	MDQC-PC
Affy. MAQC	120	(34,34,23)	(0,0,0)	(9,3,1)
Illu. MAQC	19	(0,0,0)	(0,0,0)	(3,1,0)
Digitally contaminated arrays
Source	No. of chips	Contaminated	Chips flagged
			PMVO-PC	MDQC
Affy. spike-in	12	–	none	2,8,10
		1	1	1,8
		1,2	1,2,8	1,2
		1,2,11	1,2,8,11	8,10

For negative controls, table cells give number of arrays flagged at α=0.10, 0.05, 0.01.

For positive controls, cell entries give indices of arrays contaminated or identified by various algorithms. Method labels are: PMVO-raw, for parametric multivariate outlier detection applied to raw QA features; PMVO-PC, for PMVO applied with dimension reduction to first three principal components; MDQC, for Mahalanobis distance-based algorithm of Cohen Freue et al. (2007) with the MCD estimator of covariance, applied to raw QC features; and MDQC-PC, for MDQC with the S-estimator of covariance applied on PC1–PC3 of QC features.