Table 1.
For each dataset we compare the absolute deviation in methylation between adjacent type1–type2 probe pairs (probes within 200 bp of each other), averaged over probe pairs and samples, for four different normalisation methods
| Dataset | NONE (%) | PBC (%) | SWAN (%) | BMIQ (%) | P |
|---|---|---|---|---|---|
| BT | 7.8 | 6.3 | NA | 6.2 | <10−10 |
| CL | 8.6 | 18.4 | NA | 7.2 | <10−10 |
| FFPE | 9.2 | 8.0 | 8.5 | 7.8 | <10−10 |
| FF | 8.5 | 8.1 | 7.6 | 7.3 | <10−10 |
| GBM | 9.2 | 7.6 | NA | 7.5 | <10−10 |
| TCGA | 9.4 | 7.8 | 8.3 | 7.4 | <10−10 |
| LIV | 10.0 | 6.3 | 7.4 | 6.4 | ∼1 |
| LC | 10.3 | 7.0 | 7.7 | 6.7 | <10−10 |
| BLDC | 11.0 | 8.0 | 7.9 | 7.6 | <10−10 |
| HCC | 12.0 | 8.5 | 8.7 | 8.1 | <10−10 |
NONE refers to the case of no adjustment for probe design type. The last column give the paired Wilcoxon rank sum test P-value (treating each probe-pair deviation in each sample as a separate value), assessing the statistical significance that the absolute deviation for BMIQ is smaller than the next best competing method. NA indicates non-available owing to lack of access to idat files needed for processing by SWAN. In bold-face we show the smallest deviation across methods.