Research Article
Microarray Normalization Revisited for Reproducible Breast Cancer Biomarkers
Table 6
Batch correction after RMA normalization: percent misclassification in the prediction of hormone receptors.
| | # samples for testing | no batch correction | ComBat: ViBatch: {GSE} | fSVA: ViBatch: {ER, GSE} 3014 samples for training | fSVA: ViBatch: {ER, HER2, GSE} 2291 samples for training | fSVA: ViBatch: {ER, PGR, HER2, GSE} 1825 samples for training | RmaMSingle | RmaMGlobal | RmaMSingle | RmaMSingle | RmaMGlobal | RmaMSingle | RmaMGlobal | RmaMSingle | RmaMGlobal |
| # SurrVar | | — | — | — | 9 | 8 | 5 | 9 | 8 | 10 | ESR1 | 3014 | 8.5 | 8.0 | 19.6 | 20.7 | 8.1 | 10.5 | 8.0 | 23.6 | 8.1 | PGR | 2170 | 14.1 | 12.5 | 19.0 | 31.5 | 13.2 | 24.2 | 13.4 | 27.5 | 13.3 | HER2 | 2443 | 9.6 | 8.5 | 15.1 | 11.6 | 8.7 | 10.3 | 7.0 | 24.2 | 7.1 | Average | | 10.7 | 9.7 | 17.9 | 21.3 | 10.0 | 15.0 | 9.5 | 25.1 | 9.5 |
|
|