In the face of recurrent epidemics caused by RNA viruses such as those responsible for influenza and HIV, public health agencies around the world have set up surveillance programs to uncover new emerging strains that may become a threat to human populations. However, this prediction exercise is fraught with difficulties. Here we present three approaches to address this outstanding issue. First we describe a novel posterior predictive modeling framework that aims at forecasting the emergence of influenza viruses from past seasons; despite the sophistication of the Bayesian model, its predictive power is somewhat disappointing, but the approach can be improved in several ways. Second, we resort to an ad-hoc combination of phylogenetic, time series and network analyses to estimate current transmission networks at a global scale, and show that this approach, based on a stratified design, is extremely fast and uncovers a network shift in H1N1 viruses at the emergence of the 2009 influenza pandemic. Last, in-between the complex Bayesian modeling and the ad-hoc approach, we describe a series of Approximate Bayes Computing devices to estimate current HIV transmission networks. We show that a Sequential Monte Carlo sampler outperforms the na"ive and the Markov chain Monte Carlo samplers in terms of both convergence speed and accuracy, and makes it possible to distinguish spatially structured HIV networks.
Dept of Biostatistics
biostats [at] columbia [dot] edu