Fundamental Identifiability Limits in Molecular Epidemiology

Area of Science:

• Epidemiology
• Virology
• Mathematical Biology

Background:

• Viral phylogenies are vital for tracking infectious disease spread and estimating epidemiological parameters like the effective reproduction ratio (Re).
• Phylodynamic inferences can supplement or replace traditional surveillance, especially with poor or delayed sampling.
• The robustness of phylodynamic inferences under uncertainty in pathogen prevalence and sampling intensity is not well understood.

Purpose of the Study:

• To fully characterize the information extractable from serially collected viral phylogenetic data using the birth-death-sampling model.
• To assess the reliability of phylodynamic inferences in reconstructing epidemiological dynamics.

Main Methods:

• Application of recently developed mathematical techniques to analyze viral phylogenetic data.
• Utilizing the birth-death-sampling model to represent epidemiological scenarios.
• Investigating the identifiability of epidemiological parameters from phylogenetic data.

Main Results:

• Identified numerous "congruent" epidemiological scenarios that are indistinguishable using phylogenetic data alone, regardless of dataset size.
• Demonstrated that maximum-likelihood and Bayesian inference methods cannot reliably reconstruct true epidemiological dynamics without strong constraints or rate priors.
• Showed that estimators converge to a "congruence class" of dynamics rather than the true dynamics.

Conclusions:

• Phylogenetic data alone is insufficient for robust reconstruction of viral epidemiological dynamics.
• Robust epidemiological inferences require strategies beyond standard phylogenetic fitting methods.
• Proposed concrete strategies to improve the reliability of phylodynamic inferences from viral data.