Title: Machine Learning Guided Mechanistic Model Discovery for Respiratory Infections

Abstract: The immune response to influenza infection can be highly variable. Contributors to this variability include the infecting viral strain, biological sex, and age of the patient, etc. The pandemic H1N1 strain, for example, is efficient at spreading whereas avian influenza strains like H5N1 and H7N9 are more sporadic but significantly deadlier. Moreover, reproductive aged biological females show increased severity and mortality compared to age-matched males. Finally, children are disproportionately susceptible to influenza infection compared to adults despite having similar viral loads. Taken together, the immune response to influenza infections is highly variable, with viral strain, sex, and biological age being significant contributors. Mathematical modelling provides a unique opportunity to generate hypotheses explaining the observed variability in immune response to influenza infection. This proposal aims to develop a novel machine learning-guided mechanistic modelling paradigm to generate an ensemble of models, with each model representing a hypothetical mechanism of action—capable of explaining the observed data. Drivers of sex-specificity in highly pathogenic influenza strains will be evaluated using novel murine datasets of H5N1 and H7N9 viruses. This work will be completed in two parts. Aim-1 will focus on curating publicly available datasets and aggregating them using a novel scheme. This aggregated dataset will be used to develop an ensemble of mechanistic ordinary differential equation models using a novel machine learning-based framework. The developed ensemble of models will be validated on unseen datasets to ensure validity of the approach. Aim-2 will focus on applying the novel machine learning-guided mechanistic modelling paradigm to develop an ensemble of models that explain the mechanisms driving sex-specificity in highly pathogenic influenza infections. Altogether, this proposed research will result in a modelling paradigm capable of high throughput hypothesis generation, as well as identification of several potential treatment targets for highly pathogenic influenza infection, curated for males and females.

Chair:

Dr. Jason Shoemaker

Department of Chemical and Petroleum Engineering, University of Pittsburgh

Committee Members:

Dr. John Alcorn

Department of Immunology, Pediatrics, University of Pittsburgh

Dr. Robert Parker

Department of Chemical and Petroleum Engineering, University of Pittsburgh

Dr. Ipsita Banerjee

Department of Chemical and Petroleum Engineering, University of Pittsburgh