NOTE: There has been a cancellation. Therefore, no seminar will be held today.

Join Zoom Meeting

https://ncsu.zoom.us/j/98859624475?pwd=6ei7OBEGOYXJEoMyujk1tVj4OBsVem.1

Meeting ID: 988 5962 4475
Passcode: 655269

Title and Abstract

Dynamically optimizing pathogen genomic surveillance to inform the management of antimicrobial resistance

Genomic surveillance of infectious pathogens provides many insights into their epidemiological and evolutionary dynamics not available from other sources of data. Yet it is generally unclear how to design sampling strategies that maximize the amount of epidemiological useful information when finite resources limit sampling to only a small fraction of all cases. By adopting a sequential decision making framework, we recently showed how Markov decision processes (MDPs) can be applied to optimize genomic sampling. However, this framework only considers the statistical information gained from genomic surveillance and not the value of this information to disease control efforts. While it is clear that control strategies require reliable up-to-date information, it is less clear how to optimize surveillance to inform control strategies. Using antimicrobial resistance (AMR) management as a case study, we consider how to jointly optimize both surveillance and control (i.e. treatment). This creates a challenging optimization problem because there is both a resource allocation trade-off and a trade-off between the short-term benefits of antimicrobial use to reduce near-term disease prevalence and the long-term costs of antimicrobial use due to resistance evolution. To find optimal strategies, we take advantage of reinforcement learning methods based on Deep Q-Learning to predict the long-term expected rewards (future healthy patient days) of different surveillance and treatment strategies. Learning to predict these long-term rewards allows us to find surveillance strategies that can be leveraged into improved AMR management strategies through acquiring information about the evolutionary dynamics of AMR.

Biography

David Rasmussen is an Associate Professor in the Department of Entomology and Plant Pathology at NC State. He leads the Phylodynamics Research Group, which develops new computational and statistical methods for genomic epidemiology, population genomics and phylogenetics.