Feature extraction methods for mutation screening

Abstract

DNA sequencing is the gold standard in clinical diagnosis of genetic disorders. Existing mutation screening methods heavily depends on base-calling algorithms which produces DNA sequences by analysing the peaks in the chromatogram signal. This study introduces a new method working only on signals eliminating the possible errors caused by base-calling. The method compares a patient's DNA chromatogram with a reference to discover possible mutations. First step is to transform the original signals to the frequency domain via DCT. They are next used to create a similarity matrix in which consecutive high similarity points along the main diagonal is obtained by using dynamic programming. The chromatograms are re-sampled to produce a new equal length pair by using coordinates of the path. Values of similarity matrix along the path itself corresponds to the difference of chromatograms in which possible mutations are detected by inspecting the peaks. We compared the performance of our method to a mutation screening software.