Accounting for contact network uncertainty in epidemic inferences with Approximate Bayesian Computation

Area of Science:

• Epidemiology
• Computational Biology
• Network Science

Background:

• Realistic infectious disease models require incorporating complex contact network structures.
• Existing models often assume perfect knowledge of contact networks and epidemic event times, which is unrealistic.
• Imperfect data on contact patterns and event timing introduces significant uncertainty into disease spread analysis.

Purpose of the Study:

• To develop a novel Bayesian inference framework to address uncertainties in epidemic and contact network data.
• To enable accurate parameter estimation for contagious processes despite imperfect observations.
• To apply the developed method to analyze the spread of Tattoo Skin Disease (TSD) in bottlenose dolphins.

Main Methods:

• Network-augmented Mixture Density Network-compressed Approximate Bayesian Computation (NA-MDN-ABC) is proposed.
• The method learns informative summary statistics from imperfect epidemic and network data.
• Bayesian inference is performed to estimate parameters of contagion spread.

Main Results:

• The NA-MDN-ABC method effectively handles uncertainty in contact network structure and epidemic event timing.
• Simulated epidemics and networks demonstrate the utility of the proposed framework.
• The approach is successfully extended to analyze real-world disease spread data.

Conclusions:

• The NA-MDN-ABC method provides a robust approach for Bayesian inference in infectious disease modeling with imperfect data.
• Accounting for network and observational uncertainty is crucial for accurate contagion spread analysis.
• This framework has broad applicability, including the study of marine mammal diseases like TSD.