Causal Reasoning for Predicting Viral Pathogenicity

PI: Kristie Oxford
The biodefense community needs data analysis methods that can rapidly predict the determinants of viral pathogenesis or identify targets for medical countermeasures in emerging pathogens, such as SARS-CoV-2. Currently, systems biologists use data-driven ML approaches to identify factors that are associated with pathogenicity, but these methods do not separate causes from effects. This project aims to distinguish between causal factors of pathogenesis and other correlated factors.
Researchers at PNNL are developing a causal inference engine that learns from both data and prior knowledge of host response pathways to discover causal mechanisms involved in viral pathogenesis. We expect our causal inference engine will help us predict how perturbing features individually, or in combination, could impact host responses, and therefore inform experiments to validate key markers for predictive pathogenesis and to identify plausible medical interventions. To integrate high-throughput data with the causal model, we are applying data reduction and fusion techniques to identify feature patterns that constitute molecular signatures of host response to viral infection. We then apply causal reasoning to interrogate the model with data to predict determinants of viral pathogenesis and identify putative therapeutic targets.
These capabilities will contribute to a response effort designed for the biodefense community to rapidly screen emerging pathogens and provide mechanistic explanations for why particular treatment options may be promising.