Comparing methods to estimate time-varying reproduction numbers using genomic and epidemiological data

Estimating the time-varying reproduction number $R_t$ during an epidemic is important. $R_t$ indicates whether an epidemic is growing or declining and can aid in assessing the impact of interventions. Recent advances have enhanced methods for estimating $R_t$ and other epidemiological parameters from surveillance and genomic data independently. The Birth-Death Skyline (BDSKY) in BEAST 2.5 and EpiEstim are two common methods used to estimate $R_t$ from these data sources. We introduce an outbreak simulation platform that generates pathogen sequence data and epidemiological linelists. We use this platform to assess $R_t$ estimation methods' accuracy under various sampling scenarios similar to what was observed during past epidemics. We identified biases and determined appropriate scenarios for improving the accuracy of $R_t$ estimation approaches based on multiple outbreak simulations. When data becomes sparse and unreliable, genomic sequence data provide reasonable $R_t$ estimates even when sampling is not uniform.