 |
Center for Nonlinear Studies |
Accurate classification of HIV and the detection of recombinations in HIV genomes is crucial for epidemiological monitoring and to develop potential vaccines. Recently we developed jpHMM, a probabilistic model that predicts phylogenetic recombination breakpoints in genomic sequences from HIV and assigns each position in an input sequence to one predicted parental subtype. The prediction is based on a multiple sequence alignment of the major HIV-1 subtypes. jpHMM models each subtype in the alignment as a profile HMM, but in addition transitions between different subtypes are possible. For a given input sequence, the assignment of parental subtypes is defined by the most probable path through the model, the Viterbi path. Systematic test runs demonstrate that jpHMM is more accurate then traditional recombination detection programs.
Recently, we extended jpHMM to include information about the reliability of the program output. The program now identifies regions where the model is 'uncertain' about the parental subtype, and it estimates for every breakpoint an interval in which the breakpoint can be expected with high probability. To this end, we calculate the posterior probability for each subtype at each position, i.e. the probability that the respective subtype is the true parental subtype at this position under our probabilistic model.
[1] Schultz et al., A Jumping Profile Hidden Markov Model and Applications to Recombination Sites in HIV and HCV Genomes. BMC Bioinformatics 7:265. 2006.
[2] Zhang et al., jpHMM at GOBICS: a web server to detect genomic recombinations in HIV-1. NAR, 34. 2006.
Host: Ming Zhang, T-6