Supplementary Information: A comparison of background correction methods for two-colour microarrays

Matthew E. Ritchie^a, Jeremy Silver^b, Alicia Oshlack^b, Melissa Holmes^c, Dileepa Diyagama^d, Andrew Holloway^d and Gordon K. Smyth^b

a. Department of Oncology, University of Cambridge, CRUK Cambridge Research Institute
b. Bioinformatics Division, The Walter and Eliza Hall Institute of Medical Research
c. Immunology Division, The Walter and Eliza Hall Institute of Medical Research
d. Ian Potter Foundation Centre for Cancer Genomics and Predictive Medicine, The Peter MacCallum Cancer Centre

Appendix

Further details on the different models and their implementation

Additional Plots

• Mixture experiment
  • MA plot for slide 1 (0:100 vs 100:0)
  • MA plot for slide 2 (100:0 vs 0:100)
  • MA plot for slide 3 (0:100 vs 94:6)
  • MA plot for slide 4 (94:6 vs 0:100)
  • MA plot for slide 5 (0:100 vs 88:12)
  • MA plot for slide 6 (88:12 vs 0:100)
  • MA plot for slide 7 (0:100 vs 76:24)
  • MA plot for slide 8 (76:24 vs 0:100)
  • MA plot for slide 9 (0:100 vs 50:50)
  • MA plot for slide 10 (50:50 vs 0:100)
  • MA plot for slide 11 (0:100 vs 0:100)
  • MA plot for slide 12 (0:100 vs 0:100)
  • Smoothed variances from non-linear model versus intensity for each method.
    By varying the offset k used in the normexp+offset method, we can recover an intensity dependent profile similar to that obtained for the standard method (k = -10) or the no background method (k = 190).
  • MA plot for slide 12 (0:100 vs 0:100), using normexp method with varying offset (k=25,50,75,100)
  • limma eBayes FDR plot for normexp method with varying offset (k=25,50,75,100)
• Spike experiment
  • Bias of LUS controls: M values from slide 1
  • Bias of LUS controls: M values from slide 2
  • Bias of LUS controls: M values from slide 3
  • Bias of LUS controls: M values from slide 4
  • Bias of LUS controls: M values from slide 5
  • Bias of LUS controls: M values from slide 6
  • Bias of LUS controls: M values from slide 7
  • Bias of LUS controls: M values from slide 8
  • Bias of LUS controls: M values from slide 9