Discussion - Newport iServer MicroServer iTHX-M Operator's Manual

Table D.1
Performance on HapMap dataset for DM and BRLMM at various fixed thresholds.
Method Confidence
Threshold
DM 0.26
DM 0.33
BRLMM 0.3
BRLMM 0.4
BRLMM 0.5
BRLMM 0.6

Discussion

370
Affymetrix GeneChip ®
Overall Hom
Call Rate Call
Rate
94.16% 97.24% 86.32%
95.96% 98.24% 90.16%
97.40% 97.40% 97.75%
98.27% 98.30% 98.48%
98.79% 98.82% 98.93%
99.15% 99.18% 99.25%
BRLMM provides a significant improvement over the DM method,
raising call rates and accuracy, and just as importantly, establishing
balanced performance between homozygotes and heterozygotes. As a
multiple-chip method, it has some extra considerations which need to
be taken into account in practice.
One matter to consider is the batch size in which to apply BRLMM.
While more samples will generally lead to better performance, we
have found that, for good data sets, performance reaches a plateau with
as few as 50 samples; whereas for lower-quality data sets, it can take as
many as 100. The working definition of good used here is a data set
with an average call rate of greater than or equal to 93% (for 500K)
using DM at a threshold of 0.33, and the definition of lower-quality
used here is a data set with an average call rate of less than or equal to
93% using DM at a threshold of 0.33 (these call rates increase when
called with BRLMM). Numbers of samples are actually numbers of
distinct DNAs analyzed – a data set consisting of many replicates of
the same sample may not have sufficient genetic diversity to build the
prior. Note that with the default settings, BRLMM requires at least
two observations of each genotype to build the prior, so the absolute
Genotyping Analysis Software User's Guide
Het Overall
Call Concordance
Rate
99.15%
98.94%
99.40%
99.31%
99.26%
99.17%
Hom Het
Concordance Concordance
99.39% 98.38%
99.27% 97.93%
99.34% 99.55%
99.25% 99.47%
99.20% 99.40%
99.11% 99.33%