Analytical pipelines for data and model integration: finding informed pathways for antimicrobial resistance control

Routinely-collected data generated in surveillance and diagnostic laboratories and medical records contain
information valuable for understanding the transmission of antimicrobial resistant pathogens. This information
can be harnessed to inform transmission models and to design and evaluate mitigation strategies. However,
current modeling approaches often focus on addressing one resistance on a pathogen at a time which leads to
key features of resistant pathogen dynamics such as multidrug resistance and pathogen interactions to not be
commonly addressed. In order to capitalize on the data being generated via surveillance and diagnostic
activity, we propose to carry out the following research activities: 1) we will develop graphical models to
integrate multiple data streams (phenotypic, genotypic resistances and metadata) to support analysis and
visualization of complex resistant patterns and joint distribution of resistances, 2) we will apply and evaluate
the analytical pipeline to data collected in national-level surveillance systems, 3) we will develop and evaluate
agent-based models for pathogen transmission in health-care settings that incorporate multidrug resistance
features and pathogen interactions and apply graphical modeling approaches to analyze and validate the
models, and 4) we will disseminate the tools by creating open source packages and through website
implementation. With the developed tools we will provide a path to quantify changes on complex resistance
patterns over time or across sources, identify drugs that can lead to further selection of first choice drugs,
identify cluster of risk factor for resistance and evaluate vaccination, antibiotic stewardship and heterogeneity
interventions in the presence of pathogen and resistance interactions.
The challenges of dealing with multiple streams of data and complex models are not unique to infectious
disease research. The developed workflows will be applicable to a broad spectrum of biomedical research
questions, particularly those that involve the collection of mixed data and simultaneous genotypic and
phenotypic data. Similarly, agent-based models are increasingly used across all biomedical disciplines, from
molecular biology to epidemiology, and therefore advances in their analyses can have a broader positive
impact in multiple biomedical disciplines.