Bayesian analysis of complex interacting mutations in HIV drug resistance and cross-resistance.
Ivan Kozyryev,Jing Zhang
DOI: https://doi.org/10.1007/978-94-017-9245-5_22
2015-01-01
Abstract:A successful treatment of AIDS world-wide is severely hindered by the HIV virus' drug resistance capability resulting from complicated mutation patterns of viral proteins. Such a system of mutations enables the virus to survive and reproduce despite the presence of various antiretroviral drugs by disrupting their binding capability. Although these interacting mutation patterns are extremely difficult to efficiently uncover and interpret, they contribute valuable information to personalized therapeutic regimen design. The use of Bayesian statistical modeling provides an unprecedented opportunity in the field of anti-HIV therapy to understand detailed interaction structures of drug resistant mutations. Multiple Bayesian models equipped with Markov Chain Monte Carlo (MCMC) methods have been recently proposed in this field (Zhang et al. in PNAS 107: 1321, 2010 [1]; Zhang et al. in J Proteome Sci Comput Biol 1: 2, 2012 [2]; Svicher et al. in Antiviral Res 93(1): 86-93, 2012 [3]; Svicher et al. in Antiviral Therapy 16(7): 1035-1045, 2011 [4]; Svicher et al. in Antiviral Ther 16(4): A14-A14, 2011 [5]; Svicher et al. in Antiviral Ther 16(4): A85-A85, 2011 [6]; Alteri et al. in Signature mutations in V3 and bridging sheet domain of HIV-1 gp120 HIV-1 are specifically associated with dual tropism and modulate the interaction with CCR5 N-Terminus, 2011 [7]). Probabilistically modeling mutations in the HIV-1 protease or reverse transcriptase (RT) isolated from drug-treated patients provides a powerful statistical procedure that first detects mutation combinations associated with single or multiple-drug resistance, and then infers detailed dependence structures among the interacting mutations in viral proteins (Zhang et al. in PNAS 107: 1321, 2010 [1]; Zhang et al. in J Proteome Sci Comput Biol 1: 2, 2012 [2]). Combined with molecular dynamics simulations and free energy calculations, Bayesian analysis predictions help to uncover genetic and structural mechanisms in the HIV treatment resistance. Results obtained with such stochastic methods pave the way not only for optimization of the use for existing HIV drugs, but also for the development of the new more efficient antiretroviral medicines. In this chapter we survey current challenges in the bioinformatics of anti-HIV therapy, and outline how recently emerged Bayesian methods can help with the clinical management of HIV-1 infection. We will provide a rigorous review of the Bayesian variable partition model and the recursive model selection procedure based on probability theory and mathematical data analysis techniques while highlighting real applications in HIV and HBV studies including HIV drug resistance (Zhang et al. in PNAS 107: 1321, 2010 [1]), cross-resistance (Zhang et al. in J Proteome Sci Comput Biol 1: 2, 2012 [2]), HIV coreceptor usage (Svicher et al. in Antiviral Therapy 16(7): 1035-1045, 2011 [4]; Svicher et al. in Antiviral Ther 16(4): A14-A14, 2011 [5]; Alteri et al. in Signature mutations in V3 and bridging sheet domain of HIV-1 gp120 HIV-1 are specifically associated with dual tropism and modulate the interaction with CCR5 N-Terminus, 2011 [7]), and occult HBV infection (Svicher et al. in Antiviral Res 93(1): 86-93, 2012 [3]; Svicher et al. in Antiviral Ther 16(4): A85-A85, 2011 [6]).