Empirical evaluations of analytical issues arising from predicting HLA alleles using multiple SNPs

BMC Genomic Data

Table 7 Prediction accuracies of HLA alleles with frequency exceeding a threshold

Training set		Intermediate Resolution			High Resolution
	HLA-	0.05*	0.03	0	0.05	0.03	0
Caucasians in the FHCRC cohort (N = 1280)	A	99	99	98	99	99	97
	B	98	97	96	97	97	94
	C	97	96	96	97	97	96
	DRB1	97	97	97	91	91	88
	DQB1	99	99	99	98	98	98
Caucasians in the STEP cohort (N = 832)	A	99	99	95	99	98	95
	B	97	97	95	97	97	93
	C	96	94	94	95	95	94
	DRB1	98	98	98	98	95	91

* Threshold for HLA allele frequency
The accuracies of HLA predictive models, built with a training set of Caucasian samples from the FHCRC cohort
(genotyped on Affy 5.0) or the STEP cohort (genotyped on Illumina 1.2M), were evaluated using the WTCCC samples
(genotyped on Affy 500K) as the validation set (N = 1480). Only HLA alleles with frequency more than threshold
0.05, 0.03 or 0 in the validation set were evaluated for their prediction accuracies. The confidence threshold was set at CT = 0.

ISSN: 2730-6844