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

Results

Appendix D | Bayesian Robust Linear Model with Mahalanobis (BRLMM) Distance Classifier Al-
The results in Table D.1
chips. The default BRLMM threshold is set to 0.5, and it can be tuned
for higher call rates or higher accuracy depending on which one is
more suitable for the requirements of downstream analysis.
The following is a list of performance advantages of BRLMM over
DM:
• At the default confidence of 0.5, BRLMM achieves approximately
a 2/3 reduction in the no-call rate over DM at a threshold of 0.33
– a finding that has been consistently observed on a variety of
data sets.
• BRLMM achieves an improvement of about 0.1% in HapMap
concordance, compared to DM with a more stringent threshold of
0.26 (about 0.2% compared to DM with a threshold of 0.33).
• BRLMM has a large reduction in call rate and accuracy differences
between homozygotes and heterozygotes. In the evaluation of
concordance on homozygotes, there were 25 SNPs which
exhibited what appeared to be allele swap.
• Allele Swap is when the concordance with HapMap genotypes is
extremely low (on the order of 20% or less) but jumps to very
high (90% or more) when the A and B alleles are swapped. Such
SNPs are probably due to errors in HapMap, which swaps the
alleles before reporting final accuracy.
are based on combining the Nsp and Sty