From Assessments to Decisions: How to Leverage Bayesian Networks
Citation:
Carriger, J. From Assessments to Decisions: How to Leverage Bayesian Networks. AAAS Annual Meeting 2019, DC, Washington, February 14 - 17, 2019.
Impact/Purpose:
This will be a presentation at AAAS Annual Meeting in a session on synthetic biology and adaptive management. The purpose of the presentation is to discuss how Bayesian networks can be leveraged for adaptive management and synthetic biology decisions.
Description:
Bayesian networks provide powerful and flexible modeling platforms for examining uncertainty and making inferences. They are especially useful for developing models that reflect causal issues such as how a stressor might impact an environmental endpoint of concern or how a decision can lead to an unintended consequence. The graphical component of a Bayesian network can help capture expert knowledge about the relationships among the uncertain events in a problem. This may be augmented by the transferability between Bayesian networks and conceptual models developed for risk assessments to develop a useful causal structure for assessments and decision making. The strength of the relationships between causes and effects in a Bayesian network are estimated through conditional probabilities for the uncertainties of different occurrences. Causal modeling with a Bayesian network can, therefore, be used as a platform to combine statistical and mechanistic knowledge for risk management decision-focused models. For science-based risk management processes, Bayesian networks can be a useful tool for capturing, improving, and communicating the knowledge base. As new information becomes available, the structural components of a Bayesian network can be readily updated by adding or removing variables or changing the cause and effect flows between existing variables. The probabilities can also be updated as more information is obtained, and uncertainties are resolved. These capabilities for updating what is known about a problem make Bayesian networks a powerful tool for representing the knowledge base in adaptive management. Previous adaptive management frameworks with Bayesian networks have been under-utilized but recent advances in causal modeling have continued to expand the capabilities for using Bayesian networks for adaptive management applications. Bayesian networks provide flexible modeling platforms for making inferences with available information, whether rich or poor. The impacts on predictions from uncertainties can be examined at any level of the model and scenarios can be examined and compared through flexible inferential capabilities. Causal knowledge on the components and connections and their uncertainties is the basis for powerful insights through causal pathway analysis and through causal and predictive assessments for decision analysis. Bayesian networks are very useful tools for complex scientific and environmental problems that can help unify synthetic biology risk assessment with the risk management and decision analysis process.