Manuchehr Aminian, Cal Poly Pomona
Data-driven characterization of influenza-like illnesses and predicting time of exposure
Understanding and quantifying the time-evolving dynamics between a pathogen and the host immune system is of fundamental importance toward the ongoing fight against infectious diseases. Having accurate information of the stage of a person's immune response can help identify if someone is infected — possibly even when they aren't symptomatic. Applied to larger scales, predicting how long someone has been infected with a pathogen can inform proper treatment (e.g., prescription of Oseltamivir for influenza A/B) and may be useful to improve the precision of contact tracing.
We present our results incorporating this assumption of the time-varying behavior of the immune system in familiar workflows for bioinformatics and systems biology: the joint problems of feature selection of biomarkers, and prediction of the host's state from a machine learning lens. Specifically, we consider two types of problems: (1) early prediction of a binary infected variable within 24 hours of exposure; (2) prediction of time of exposure to pathogen, with the knowledge a person is already infected. We will discuss the results of our computational experiments when applying this to microarray transcriptomic data from an ensemble of human challenge studies (GSE73072) concerning influenza-like illnesses.