Identification of heterozygous point mutation events in DNA sequencing chromatograms

The recent discovery of activating somatic mutations in cancer that correlate with phenotypes such as drug responsiveness, has generated renewed interest in the sequencing of genomes of tumor samples and cancer cancer cell lines with the goal of identifying the set of mutations that produce those phenotypes [1]. The two most popular strategies for discovering these events are array CGH [2], and direct sequencing of tumor samples and cells at specific loci of genes suspected a priori to be involved in tumor proliferation and survival. The latter technique involves using PCR amplification of the loci of interest and standard capillary electrophoresis DNA sequencing to generate chromatograms and sequences which are then compared to a reference normal sequence to reveal mutations. The detection of homozygous events is relatively straightforward, but the identification of heterozygous point events is problematic. The process of detecting heterozygous events involves detecting "peaks within peaks" of chromatogram waveforms and is plagued by a variety of artifacts in these signals which can potentially generate false positives. Proposed herein is a detection algorithm based on a classifier which distinguishes candidate "peaks within peaks" that are heterozygous point mutations from those that are false positives based on statistics about the candidate event and representation of these artifacts as interval-scale input variables to a machine learning algorithm.