Skip to main content

Advertisement

Table 1 Prediction results for feature ranking methods (chi squared, SWRF, SLR and LP) on two LOAD datasets (TGen and ADRC) with two sets of SNPs (chromosome 19 only and genome-wide)

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

Dataset # SNPs Method Number of SNPs used in classifier
1 2 5 10 50 100 500 1000
TGen 3,652 (chr19) Chi Sq 0.6628 ± 0.0284 0.6791 ± 0.0278 0.6811 ± 0.0280 0.6567 ± 0.0290 0.5768 ± 0.0306 0.5925 ± 0.0300 0.5627 ± 0.0306 0.5512 ± 0.0304
SWRF 0.6783 ± 0.0284 0.6778 ± 0.0284 0.6791 ± 0.0282 0.6697 ± 0.0286 0.6620 ± 0.0286 0.6280 ± 0.0290 0.5941 ± 0.0304 0.5691 ± 0.0306
SLR 0.5014 ± 0.0302 0.6821 ± 0.0282 0.7129 ± 0.0270 0.7170 ± 0.0269 0.6847 ± 0.0282 0.6747 ± 0.0286 * *
LP 0.6733 ± 0.0284 0.6904 ± 0.0278 0.7080 ± 0.0270 0.7184 ± 0.0267 0.7093 ± 0.0270 0.6945 ± 0.0274 0.6246 ± 0.0298 0.5989 ± 0.0302
234,665 (chr1-22) Chi Sq 0.6628 ± 0.0284 0.6991 ± 0.0274 0.7230 ± 0.0267 0.7310 ± 0.0263 0.7068 ± 0.0272 0.6549 ± 0.0292 0.6059 ± 0.0302 0.5990 ± 0.0302
SWRF 0.6640 ± 0.0284 0.6705 ± 0.0284 0.7020 ± 0.0280 0.6796 ± 0.0286 0.6749 ± 0.0284 0.6087 ± 0.0306 0.5447 ± 0.0253 0.5261 ± 0.0169
SLR 0.6783 ± 0.0284 0.7076 ± 0.0270 0.7291 ± 0.0261 0.7424 ± 0.0257 0.7464 ± 0.0257 * * *
LP 0.6733 ± 0.0284 0.6904 ± 0.0284 0.7088 ± 0.0269 0.7396 ± 0.0257 0.7519 ± 0.0251 0.7286 ± 0.0270 0.6138 ± 0.0237 0.5735 ± 0.0178
ADRC 13,087 (chr19) Chi Sq 0.6834 ± 0.0220 0.7369 ± 0.0206 0.7433 ± 0.0204 0.7169 ± 0.0212 0.6446 ± 0.0229 0.6109 ± 0.0233 0.5282 ± 0.0239 0.5361 ± 0.0241
SWRF 0.6834 ± 0.0229 0.7006 ± 0.0221 0.7169 ± 0.0206 0.7122 ± 0.0206 0.6894 ± 0.0220 0.6580 ± 0.0225 0.5343 ± 0.0235 0.4965 ± 0.0239
SLR 0.6834 ± 0.0220 0.6855 ± 0.0218 0.6964 ± 0.0216 0.7068 ± 0.0213 0.7041 ± 0.0213 0.6478 ± 0.0227 * *
LP 0.6325 ± 0.0220 0.6756 ± 0.0214 0.7342 ± 0.0210 0.7378 ± 0.0212 0.6894 ± 0.0218 0.6616 ± 0.0225 0.6095 ± 0.0241 0.5687 ± 0.0239
682,685 (chr1-22) Chi Sq 0.6834 ± 0.0220 0.7369 ± 0.0206 0.7433 ± 0.0204 0.7184 ± 0.0212 0.6438 ± 0.0227 0.6034 ± 0.0235 0.5445 ± 0.0239 0.5349 ± 0.0239
SWRF 0.6834 ± 0.0220 0.7006 ± 0.0213 0.6978 ± 0.0216 0.6934 ± 0.0220 0.6851 ± 0.0220 0.6293 ± 0.0231 0.5160 ± 0.0178 0.5029 ± 0.0127
SLR 0.6834 ± 0.0220 0.6911 ± 0.0218 0.7100 ± 0.0214 0.7354 ± 0.0206 0.6970 ± 0.0218 0.6874 ± 0.0220 * *
LP 0.6325 ± 0.0229 0.6756 ± 0.0221 0.7342 ± 0.0206 0.7315 ± 0.0206 0.7151 ± 0.0210 0.7145 ± 0.0214 0.6096 ± 0.0204 0.5435 ± 0.0122
  1. The entries are the cross-fold classification AUCs and the 95% confidence intervals obtained from application of the kNN classifier to a specified number of top-ranked SNPs. Bold cells indicate where LP significantly outperforms at least one of the other methods. The SLR method is implicitly feature-selective, reducing the feature space to under 500 features for all experiments (indicated by cells containing *).