Table 1. Summary of variability and aggregated correlation values for different within- and cross-platform normalisation.
Cross-platform | Within-platform | |||||||||||
Standardi-sation | ComBat | XPN | PMOnly | Loess Only | PMLoess | |||||||
SC | DC | SC | DC | SC | DC | SC | DC | SC | DC | SC | DC | |
Variability between replicates | ↑ with PML; ↓ with PM & L | ↑ with PM & PML; ↓ with L | ↑ with PM & L; ↓ with PML | ↑ with L; ↓ with PM & PML | ↑ with PM & L; ↓ with PML | ↓ | ↑ | ↑ | ↓ | ↓ | ↓ | ↓ |
Amplitude of noise frequencies | ↑ with PM & PML; ↓ with L | ↑ with L; ↓ with PM & PML | ↑ with L; ↓ with PM & PML | ↑ | ↓ | ↓ | ||||||
Number of highly correlated genes | ↑ | ↑ | ↑ | ↑ | ↓ | ↓ | ↑ | ↑ in PL & PS; ↓ in S | ↓ | ↓ | ↑ | ↑ in S; ↓ in PS & PL |
Average absolute correlation | ↑ | ↑ | ↑ | ↑ | ↓ | ↓ | ↑ | ↓ | ↓ | ↑ | ↑ | ↑ |
SC and DC identify results for single- and dual-channel datasets; PL, PS and S represent the three dual-channel datasets (PramilaL, PramilaS and Spellman), while PM, PML, L stand for PMOnly, PMLoess and LoessOnly, respectively. Arrows indicate whether variability and correlations are increased (↑) or decreased (↓) relative to the other normalisation procedures in the same category (cross- or within-platform).