Assessment of genomic variant calling methods through simulations

Abstract

DNA sequencing has brought many important improvements to medicine, and is a field of active development. Through the study of genetic information, we gain knowledge of hereditary diseases and traits, both in humans and other species. However, the process of sequencing is difficult, both due to vast amounts of data and sequencing errors. Verification of results is often done with expensive re-sequencing and analysis. In this thesis we study the use of simulated reads in order to obtain exact results. First we suggest a method for creating a known artificial genome, using dbSNP to provide variation. Several existing programs for variant calling are evaluated through detailed analysis of variant files. Finally we suggest methods for improving verification of results. The results show that the GATK variant callers performed well, but also Dindel provided some advantages. Furthermore, results suggest that some issues are caused by erroneous mapping and realignment. We hope that others can use these results to improve the development of sequencing algorithms through simulations.