Deep learning from phylogenies to uncover the epidemiological dynamics of outbreaks.

Widely applicable, accurate and fast inference methods in phylodynamics are needed to fully profit from the richness of genetic data in uncovering the dynamics of epidemics. Standard methods, including maximum-likelihood and Bayesian approaches, generally rely on complex mathematical formulae and approximations, and do not scale with dataset size. We develop a likelihood-free, simulation-based approach, which combines deep learning with (1) a large set of summary statistics measured on phylogenies or (2) a complete and compact representation of trees, which avoids potential limitations of summary statistics and applies to any phylodynamics model. Our method enables both model selection and estimation of epidemiological parameters from very large phylogenies. We demonstrate its speed and accuracy on simulated data, where it performs better than the state-of-the-art methods. To illustrate its applicability, we assess the dynamics induced by superspreading individuals in an HIV dataset of men-having-sex-with-men in Zurich. Our tool PhyloDeep is available on github.com/evolbioinfo/phylodeep .

Statistical modeling alongside observational data predicts long-term immunogenicity of one dose and two doses of pediatric hepatitis A vaccine in the Mendoza province of Argentina.

BACKGROUND: Follow-up for anti-hepatitis A (HA) antibody persistence up to 10 years was conducted after implementation of universal vaccination against HA virus (HAV) in Mendoza, Argentina. Based on these data, statistical modeling was used to predict the antibody persistence to 30 years. METHODS: A non-interventional study evaluated long-term immunogenicity (geometric mean concentrations [GMCs] and seroprotection rate) following routine vaccination with 1 dose (Group 1: N = 436) or 2 doses (Group 2: N = 108) of HA vaccine. Associated statistical modeling based on a Bayesian approach of mixed effects models on log transformed titers evaluated three models (linear, piecewise linear, and exponential decay, with and without a natural boosting effect). RESULTS: From the initial cohort, 9 participants (Group 1) and 1 participant (Group 2) showed antibody titers below the seroprotective threshold and received a booster. At Year 10, 190 (Group 1) and 51 (Group 2) participants remained in the study without a booster dose and all were seroprotected. Regarding statistical modeling, the piecewise linear model showed the best fit and demonstrated high and similar seroprotection for each schedule up to 30 years (89% [1-dose schedule], 85% [2-dose schedule]). The 2-dose schedule showed higher GMC (95% CI) than the 1-dose schedule (Year 10: 352 [271-456] versus 78 [69.8-87.6] mIU/mL) and Year 30 (predicted) (37 [13-97] versus 19 [11-34] mIU/mL). Natural boosting had little impact on predicted seroprotection rates at 30 years for the 1-dose schedule (89% [0.8-0.96] and 84% [0.73-0.94] with and without a natural booster, respectively). CONCLUSIONS: Long-term persistence of anti-HAV antibodies was observed up to 10 years with 1-dose and 2-dose vaccine schedules, supporting booster flexibility. Statistical modeling predicted good persistence of seroprotection for each schedule up to 30 years. Natural boosting had a limited impact on seroprotection rate predictions, enabling extrapolation of these results to non-endemic settings for traveler vaccination.