Inferring epidemiological parameters from phylogenies using regression-ABC: A comparative study

• DOI: 10.1371/journal.pcbi.1005416
• Publication Date: 2017-03-06T18:34:21Z
• Authors: Emma Saulnier; Olivier Gascuel; Samuel Alizon
• Keywords: Approximate Bayesian Computation; Lasso

Inferring epidemiological parameters such as the R0 from time-scaled phylogenies is a timely challenge. Most current approaches rely on likelihood functions, which raise specific issues that range computing these functions to finding their maxima numerically. Here, we present new regression-based Approximate Bayesian Computation (ABC) approach, base large variety of summary statistics intended capture information contained in phylogeny and its corresponding lineage-through-time plot. The regression step involves Least Absolute Shrinkage Selection Operator (LASSO) method, robust machine learning technique. It allows us readily deal with number statistics, while avoiding resorting Markov Chain Monte Carlo (MCMC) techniques. To compare our approach existing ones, simulated target trees under models settings, inferred interest using same priors. We found that, for phylogenies, accuracy regression-ABC comparable likelihood-based involving birth-death processes implemented BEAST2. Our even outperformed when inferring host population size Susceptible-Infected-Removed model. also clearly recent kernel-ABC assuming Susceptible-Infected model two types. Lastly, by re-analyzing data early stages Ebola epidemic Sierra Leone, showed provides more realistic estimates duration (latency infectiousness) than method. Overall, ABC based method able perform variable selection avoid overfitting promising analyze phylogenies.