Improvement of the jpHMM approach to recombination detection in viral genomes and its application to HIV and HBV

Accurate virus genotyping and the detection of recombinant strains are of crucial importance for understanding viral evolution as well as the design of potential vaccines and treatment strategies. A very accurate tool for detecting recombinations in genomic HIV-1 sequences is jpHMM (jumping profile Hidden Markov Model). For a given sequence, it predicts recombination breakpoints and assigns a parental subtype to each segment in between two breakpoints. In this thesis, modifications and extensions of jpHMM are carried out to improve the reliability of the recombination prediction, to reduce the runtime of the program and to allow the analysis of recombinations in circular genomes. As incorrect subtype assignments or recombination predictions may lead to wrong conclusions in epidemiological or vaccine research, it is important to assess the reliability of the predicted recombination in a particular sequence. For this reason, the output of jpHMM is extended to include a tagging of regions where the model is uncertain about the predicted subtype and an interval estimate for each predicted breakpoint. It is shown that this extension strongly improves the reliability of the recombination prediction. To allow an efficient application of jpHMM to large data sets or species with a large number of subtypes, the complex architecture of the model is substantially modified. Evaluation on HIV-1 as well as hepatitis B virus (HBV) data shows that these modifications lead to a considerable reduction of the runtime of the program. Furthermore, an extension of jpHMM to detect recombinations in viruses with circular genomes such as HBV is introduced. Recombination analysis in circular genomes is usually done on artificially linearized sequences using linear models. Since these models are normally unable to model dependencies between nucleotides at the 5’ and 3’ end of a sequence, this can result in inaccurate predictions of breakpoints and thus in incorrect classifications of circular genomes. In contrast, the circular jpHMM takes into account the circularity of the genome. Its accuracy is evaluated on a large set of recombinant HBV sequences. Additionally, about 3000 full-length HBV sequences are studied to detect socalled circulating recombinant forms (CRF). For this, certain criteria for classifying recombinant HBV sequences are proposed. Based on these criteria, 17 CRFs can be identified.