The application of network label propagation to rank biomarkers in genome-wide Alzheimer’s data

BMC Genomics

Table 2 Prediction AUCs for cross-dataset experiments

Dataset	# SNPs	Method	Number of SNPs used in classifier
Dataset	# SNPs	Method	1	2	5	10	50	100	500	1000
TGen (Feature selection from ADRC)	64,984 (ADRC overlap, chr1-22)	Chi Sq	0.6086 ± 0.0294	0.6863 ± 0.0280	0.7099 ± 0.0270	0.6958 ± 0.0253	0.6563 ± 0.0286	0.6097 ± 0.0296	0.5593 ± 0.0310	0.5563 ± 0.0308
		SWRF	0.5952 ± 0.0296	0.6980 ± 0.0274	0.6994 ± 0.0272	0.7005 ± 0.0274	0.6756 ± 0.0284	0.6677 ± 0.0284	0.5635 ± 0.0306	0.5195 ± 0.0310
		SLR	0.6086 ± 0.0294	0.6863 ± 0.0280	0.7164 ± 0.0269	0.7289 ± 0.0263	0.6522 ± 0.0292	0.6084 ± 0.0300	*	*
		LP	0.5023 ± 0.0306	0.6039 ± 0.0300	0.7023 ± 0.0272	0.7037 ± 0.0274	0.6888 ± 0.0276	0.6543 ± 0.0286	0.6114 ± 0.0298	0.5690 ± 0.0306
ADRC (Feature selection from TGen)	64,984 (TGen overlap, chr1-22)	Chi Sq	0.6172 ± 0.0231	0.6385 ± 0.0229	0.7419 ± 0.0204	0.7362 ± 0.0208	0.6695 ± 0.0225	0.6479 ± 0.0227	0.5396 ± 0.0239	0.5259 ± 0.0122
		SWRF	0.5397 ± 0.0239	0.5345 ± 0.0241	0.5350 ± 0.0241	0.5401 ± 0.0243	0.5042 ± 0.0243	0.5257 ± 0.0241	0.5201 ± 0.0241	0.5053 ± 0.0241
		SLR	0.5397 ± 0.0214	0.7006 ± 0.0214	0.7003 ± 0.0214	0.7048 ± 0.0216	0.6048 ± 0.0233	0.5854 ± 0.0237	*	*
		LP	0.5397 ± 0.0239	0.6021 ± 0.0235	0.7283 ± 0.0210	0.7366 ± 0.0208	0.6853 ± 0.0220	0.6598 ± 0.0225	0.5678 ± 0.0239	0.5306 ± 0.0239

The entries are the cross-dataset classification AUCs and the 95% confidence intervals. Feature selection was applied to one dataset, and the top-ranked features were used to derive and evaluate a kNN classifier on the other dataset. The SLR method is implicitly feature-selective, reducing the feature space to under 500 features for all experiments (indicated by cells containing *).

ISSN: 1471-2164