Normalization of Affymetrix DNA chip

D. Puthier

Inserm U1090/TAGC

Press F11 for full screen

Affymetrix DNA Chip

Technology

• In situ synthesis of oligonucleotides.
• Features.
• Cells: 24µm x 24µm
• ~ 1e7 oligos per cell
• ~ 4e5-1.5e6 probes per chip

Affymetrix DNA Chip

Probes

• Short oligonucleotides (25 mers)
• 3' arrays (3' IVT Expression Analysis)


• PM and MM
• 10-20 probes/gene


• Exon/Gene arrays (Whole-Transcript Expression Analysis)
• No more MM
• ~4 probes per exon, ~40 per gene (Exon Arrays)
• ~26 probes per gene (Gene arrays)

Affymetrix DNA Chip

Mismatch and perfect match

• Perfect Match (PM): specific probe
• Mismatch (MM): degenerated probe
• ProbePair: a pair of PM and MM
• ProbeSet: a set of x PM and MM targeting a gene (e.g; U12140_at)

Normalization: main sources of artefact

• RNA quantity / quality
• Labelling efficiency
• Scanner parameters
• Local background

Normalizing columns of a numeric matrix

Standardization (set mean to 0 and standard deviation to 1 in each column)

• centering: substracting the mean from each value
• scaling: dividing the centered value by the standard deviation

• \(\ z \) is the standardized value (the z-score)
• \(\ m_{est} \) is a sample-based estimate of the population mean
• \(\ s_{est} \) is a sample-based estimate of the population standard deviation

Normalizing columns of a numeric matrix

Robust standardization (set median to 0 and median absolute deviation to 1 in each column)

• centering: substracting the median from each value
• scaling: dividing the centered value by the median absolute deviation

• \(\ z \) is the standardized value
• \(\ \tilde{m}_{est} \) is a sample-based estimate of the population median
• \(\ \tilde{s}_{est} \) is a sample-based estimate of the population mad

Normalizing columns of a numeric matrix

Quantile normalization

• Force columns to contain the same set of values

The RMA algorithm

Robust Multichip Average

The most popular algorithm for Affymetrix data normalization

• MM values are excluded from analysis
• Global background estimation and substraction
• Quantile normalization (probe level)
• Median Polish (summarization)

The RMA algorithm

Median polish algorithm

Motivations

• Probes may display higher or lower affinity for targets

The RMA algorithm

Median polish algorithm

Procedure

• Compute a residual matrix by iteratively
• Substract the corresponding median from each row
• Substract the corresponding median from each column
• Stop if row medians and column medians are equal to zero
• Subtract the residual matrix from the initial matrix
• Compute the mean of each column to get summarized expression values

Alternatives to RMA

Other algorithms

• GCRMA
• PLIER
• RefRMA
...

Performances

• Affycomp