Using population data for assessing next-generation sequencing performance

Motivation: During the past 4 years, whole-exome sequencing has become a standard tool for finding rare variants causing Mendelian disorders. In that time, there has also been a proliferation of both sequencing platforms and approaches to analyse their output. This requires approaches to assess the performance of different methods. Traditionally, criteria such as comparison with microarray data or a number of known polymorphic sites have been used. Here we expand such approaches, developing a maximum likelihood framework and using it to estimate the sensitivity and specificity of whole-exome sequencing data. Results: Using whole-exome sequencing data for a panel of 19 individuals, we show that estimated sensitivity and specificity are similar to those calculated using microarray data as a reference. We explore the effect of frequency misspecification arising from using an inappropriately selected population and find that, although the estimates are affected, the rankings across procedures remain the same. Availability and implementation: An implementation using Perl and R can be found at busso.ncl.ac.uk (Username: igm101; Password: Z1z1nts). Contact: Darren.Houniet@ogt.com; mauro.santibanez-koref@newcastle.ac.uk