Novel algorithms for accurate DNA base-calling

Abstract

The ability to decipher the genetic code of different species would lead to significant future scientific achievements in important areas, including medicine and agriculture. The importance of DNA sequencing necessitated a need for efficient automation of identification of base sequences from traces generated by existing sequencing machines, a process referred to as DNA base-calling. In this paper, a pattern recognition technique was adopted to minimize the inaccuracy in DNA base-calling. Two new frameworks using Artificial Neural Networks and Polynomial Classifiers are proposed to model electropherogram traces belonging to Homo sapiens , Saccharomyces mikatae and Drosophila melanogaster . De-correlation, de-convolution and normalization were implemented as part of the pre-processing stage employed to minimize data imperfections attributed to the nature of the chemical reactions involved in DNA sequencing. Discriminative features that characterize each chromatogram trace were subsequently extracted and subjected to the chosen classifiers to categorize the events to their respective base classes. The models are trained such that they are not restricted to a specific species or to a specific chemical procedure of sequencing. The base- calling accuracy achieved is compared with the exist- ing standards, PHRED (Phil’s Read Editor) and ABI (Applied Biosystems, version2.1.1) KB base-callers in terms of deletion, insertion and substitution errors. Experimental evidence indicates that the proposed models achieve a higher base-calling accuracy when compared to PHRED and a comparable performance when compared to ABI. The results obtained demon- strate the potential of the proposed models for efficient and accurate DNA base-cal ling.