Computational methods for fast and accurate dna fragment assembly

As advances in technology result in the production of increasing amounts of DNA sequencing data in decreasing amounts of time, it is imperative that computational methods are developed that allow data analysis to keep pace. In this dissertation, I present methods that improve the speed and accuracy of DNA fragment assembly. One critical characteristic of automatic methods for fragment assembly is that they must be accurate. Currently, to ensure accurate sequences, the data that underlies questionable base calls must be examined by human editors so that the correct base call can be determined. This manual process is both error-prone and time-consuming. Automatic methods that yield high accuracy and few questionable calls can reduce errors and lessen the need for manual inspections. In my work, I developed a method, Trace-Evidence , that automatically produces highly accurate consensus sequences, even with few aligned sequences. Most assembly programs analyze only base calls when determining a consensus equence. The key to the high accuracy is that I incorporate morphological information about the underlying ABI trace data. This is accomplished through a new representation of traces, TraceClass, that characterizes the height and shape of traces. The new representation ot only yields high accuracy when used in consensus-calling methods, but also produces improved results when used in removing poor-quality data, and when used as inputs for neural networks for consensus determination. The need for fast processing is becoming more important as the size of sequencing projects increases. Almost all existing fragment assembly programs perform pairwise comparisons of ii