Newport iServer MicroServer iTHX-M, Operator'S Manual

Page: 364 / 414

356
Affymetrix GeneChip
®
Genotyping Analysis Software User’s Guide
Introduction
Highly accurate and reliable genotype calling is an essential
component of any high-throughput SNP genotyping technology. The
Dynamic Model (DM) algorithm
[2] , which has been extensively used
with the Affymetrix GeneChip
®
Human Mapping 100K Array Set and
the Affymetrix GeneChip ® Human Mapping 500K Array Set, has
proven to be very effective. However, Rabbee and Speed [1] recently
developed a model called the Robust Linear Model with
Mahalanobis distance classifier (RLMM – pronounced Realm ), which
provides an improvement over DM on the 100K array set with more
accurate genotype calling.
An extension of RLMM was developed for the mapping arrays and is
called
Bayesian Robust Linear Model with Mahalanobis distance
classifier (BRLMM – pronounced B-Realm) . The principal difference
between RLMM and BRLMM is the addition of a Bayesian
classification step which allows for adaptive estimation of cluster
centers and variances, avoiding RLMM’s reliance on predetermined
cluster centers and variances from known reference genotypes.
Performance improvement is achieved by two main advances over the
DM model:
• RLMM (and therefore BRLMM) performs a multiple-chip
analysis, enabling the simultaneous estimation of probe effects
and allele signals for each SNP. Accounting for probe specific
effects results in lower variance on allele signal estimates, just as it
has in the mature field of probe-level expression analysis.
• RLMM and BRLMM’s multiple-sample classification estimates
genotypes by borrowing information, as necessary, from other
SNPs to better predict the properties of the underlying clusters
corresponding to the {AA, AB, BB} genotypes. By contrast, the
DM approach calls genotypes by analyzing the probe-level
intensities one SNP and one chip at a time, using strong
assumptions about what the underlying probe intensity patterns
should look like in the context of each of the genotypes. RLMM
and BRLMM make weaker assumptions about the behavior of
GTYPE_manual.book Page 356 Wednesday, August 2, 2006 11:31 AM