#27 · Systems biology

Assumption-sensitive calibration of epidemic and phylodynamic models

The dominant thread is calibration of explicit mechanistic epidemiological and phylodynamic models against data that are not simple scalar time series, including transmission-tree forests and viral sequence datasets. Likelihood-free inference recurs: ABC-SMC is presented as a practical PyMC tutorial for ecological and epidemiological dynamical systems, and Neural Posterior Estimation is evaluated on Ebola epidemiological and phylodynamic models. Other items examine how structural choices propagate into inference. A benchmark of fixed-tree-topology strategies on viral datasets finds substitution-model parameters robust, in contrast to clock rates and node ages. Bayesian model averaging over parametric coalescent growth models addresses structural uncertainty directly. A chi-square and PERMANOVA framework compares transmission-tree epidemic forests, with PERMANOVA more sensitive. The new element is explicit attention to which calibrated quantities depend on modeling assumptions, rather than reporting a single fitted parameter set. The pattern is not uniform across items; the forest comparison and the tutorial address different questions.