Combining State Space Models and Statistical Decision Theory forPrecision Profiling of Thalassemia
Pillar developed a Bayesian caller inside PiVAT® that pairs a state space model for copy number estimation with statistical decision theory to identify alpha thalassemia subtypes from the inheritReveal™ Thalassemia research-use panel. Thirteen normal controls and 44 alpha thalassemia samples were expanded to 220 sample/normal pairs by four-fold evaluation and sequenced on the Illumina MiSeq at roughly 2,800 read pairs per amplicon. The new caller reached 88% annotation accuracy with 96% sensitivity and 93% specificity, rising to 93.7% accuracy once its own quality metric filtered poor samples, against 67% for the previous PiVAT® caller.