Objective:

To estimate infectious disease transmission patterns using a neural network-based method from case counts and pathogen sequence data.

Approach:

• Neural Posterior Estimation (NPE): A simulation-based inference method that trains a neural network on outbreak simulations to estimate epidemiological parameters.

Key Findings:

• NPE estimates closely matched those produced by Markov chain Monte Carlo (MCMC) methods.
• The method produced similar estimates to a regression-based ABC model when analyzing full-length Ebola virus sequences.
• NPE workflow was significantly faster than MCMC, taking less than 3 hours compared to approximately 11 hours.

Interpretation:

The method demonstrates how pathogen sequencing and surveillance data can inform outbreak dynamics estimates, but it is not a replacement for diagnostic testing or epidemiological investigation.

Limitations:

• The study was retrospective and assessed simple models using historical Ebola data.
• NPE may yield unreliable estimates if observed data differ significantly from training simulations.
• Model selection, neural network design, and phylogenetic reconstruction uncertainty require careful evaluation.

Conclusion:

NPE provides a framework for analyzing outbreak characteristics but relies on the accuracy of input data and disease models.

Sources:

• Proceedings of the Royal Society B