Methodological evaluation reveals substantial bias in time-dependent parameters from fixed tree topologies across viral datasets, highlighting the need for time-aware joint inference.
Key Points
Systematically evaluate the impact of using fixed tree topologies versus fully Bayesian joint estimation on phylogenetic and phylodynamic parameter estimates.
Compared unconstrained Bayesian joint inference against fixed-topology strategies conditioned on maximum likelihood trees dated using LSD or TreeTime.
Tested performance across a diverse range of viral datasets, tracking changes in substitution models, clock rates, node ages, and demographic trajectories.
Global substitution and site model parameters are largely robust to fixed-topology assumptions.
Time-dependent parameters—including molecular clock rates, node ages, and demographic history reconstructions—exhibit substantial systematic estimation discrepancies when topology is fixed.