Normalization And Allele Summarization - Newport iServer MicroServer iTHX-M Operator's Manual

Normalization and Allele Summarization

MULTICHIP ROBUST LINEAR MODEL
Appendix D | Bayesian Robust Linear Model with Mahalanobis (BRLMM) Distance Classifier Al-
The normalization and allele summarization portion of the BRLMM
algorithm consists of producing a summary value for each allele of a
SNP in each experiment. The A allele summary value increases and
decreases with the quantity of the A allele in the target genome.
Similarly, the B allele summary value increases and decreases with the
quantity of the B allele in the target genome. These summary values
are calculated to remove extraneous effects — chip-to-chip variation,
background, and the relative brightness of different probes on the
array. This approach is similar to what is used on expression arrays. For
more information, see the whitepaper, BRLMM: an Improved Genotype
Calling Method for the GeneChip
www.affymetrix.com.
First, we preprocess the data by applying quantile normalization to the
probe intensities [3], in order to minimize chip-to-chip nonbiological
variability. Normalization is essential for implementing a multi chip
model to the probe intensities. This normalization method assumes
the same underlying distribution of intensities across chips.
Second, we log2 transform the normalized intensities and robustly fit
a linear model to estimate the chip and probe effects. The details and
benefits of the Robust Multichip Average (RMA) model of probe
intensity measures have been discussed in Bolstad's paper on
normalization methods [3]. Let I denote the total number of chips
present either in the training or test sample and J denote the number
of allele A or allele B perfect-match probe intensities in the data set.
For SNP n, the model we fit to the allele A probe intensities is:
+
n n
log (y ) = S
2 Aij Ai
where y n is the normalized probe intensity for chip i, allele A probe
Aij
j and SNP n, and S
Ai
n
ß is the probe effect, and e
Aj
assumed independent, identically distributed. The probe effects are
®
Human Mapping 500K Array Set, at
+
n n
ß e where i=1,...,I; j=1,...,J
Aj Aij
n
is the chip effect determined from the A probes,
n
is an error term with mean zero,
Aij