(184b) Machine Learning and Mechanism-Based Mathematical Modeling to Identify Biomarkers and Mechanisms behind Severe Pediatric Influenza Infection | AIChE

(184b) Machine Learning and Mechanism-Based Mathematical Modeling to Identify Biomarkers and Mechanisms behind Severe Pediatric Influenza Infection


Luciani, L. - Presenter, University of Pittsburgh
Shoemaker, J. E., University of Pittsburgh
Alcorn, J., UPMC Children's Hospital of Pittsburgh
Huckestein, B. R., UPMC Children's Hospital of Pittsburgh
Pneumonia a leading cause of death in children and is responsible for 14% of all deaths of children under the age of five years[1]. A leading cause of pneumonia, influenza virus can cause anywhere between 290,000 and 650,000 deaths worldwide each year[2]. While being under the age of 6 months and medical co-morbidities are recognized risk factors for influenza associated mortality, the majority of influenza associated pediatric deaths since 2004 have occurred in older children (median age of 7) with 40% having no known co-morbidities[3]. The immense public health burden of pneumonia and influenza infection emphasizes the importance of identifying reliable and validated biomarkers that can predict severe respiratory disease in children. While potential chemokine and cytokine biomarkers for severe influenza infection have been proposed, these biomarkers are based on data derived from adults infected with 2009 H1N1 pandemic influenza. Several studies have shown that the immune responses of children to influenza infection is vastly different than those of adults, with children relying more heavily on their innate immune responses and producing lower levels of type 1 and 2 interferons (INF) compared to adults[4],[5]. Thus, there is a need for further investigation in pediatric populations. An improved understanding of the underlying mechanisms that drive severe disease can reveal pathways critical to inflammation induced lung injury allowing for improved therapeutics. Computational methods have emerged as a critical tool for revealing underlying immune response mechanisms of action, identifying key biomarkers and drug targets, and predicting complex behavior of biological systems[6]. Unfortunately, there is a lack of computational models of host driven immune responses to influenza infection in pediatric settings creating the need for a comprehensive mathematical model of pediatric immune response to influenza infection.

The overall goal of this project is to identify the underlying mechanisms and biomarkers associated with influenza severity in genetically diverse juvenile mice. In order to accomplish this objective, we apply machine learning algorithms such as Support Vector Machines (SVM) to blood and lung cytokine data, as well as flow cytometry inflammatory cell data, from juvenile mice to assess how well these biomarkers can distinguish between mild and severe disease outcomes. We hypothesize that pro-inflammatory markers, such as interleukin (IL) – 6, will be identified as key biomarkers since previous experimental studies have revealed a significant correlation between elevated levels of pro-inflammatory cytokines and severe influenza infection in both children and adults[7]. Biomarker predictions will be assessed using the receiver operating characteristic curve which assesses SVM’s ability to correctly classify juvenile mice into mild or severe outcome groups for different levels of prediction confidence.

While the ML will reveal new biomarkers to predict disease outcome, mechanisms-based mathematical modeling approaches are necessary to identifying the underlying mechanisms that promote severe infection. Non-linear ordinary differential equations (ODE) models are highly effective at characterizing complex biological systems and have been used to describe the dynamics of innate immune responses to influenza infection. Unfortunately, to our knowledge, there are no current ODE models that describe the differences of the immune response of juvenile mice that experience different levels of disease severity. We will use an ODE model trained to mouse data to determine the extent to which differences in initial cell counts and/or interaction rates contribute to immune response differences of mild and severe disease groups. After training the model to various cohorts (male/female, mild/severe outcome), we will conduct in silico experiments to determine how to best intervene to reduce viral titers and inflammation, thus improving future therapeutics.


[1] Pneumonia. (2021). Retrieved April 08, 2022, from https://www.who.int/news-room/fact-sheets/detail/pneumonia

[2] Influenza (seasonal). (2018). Retrieved April 08, 2022, from https://www.who.int/news-room/fact-sheets/detail/influenza-(seasonal)

[3] Wong KK, et al. (2013) Influenza-associated pediatric deaths in the

United States, 2004-2012. Pediatrics 132: 796-804.

[4] Levy, O., Innate immunity of the newborn: basic mechanisms and clinical correlates. Nature Reviews Immunology, 2007. 7(5): p. 379-90.

[5] You, D., et al., Inchoate CD8+ T cell responses in neonatal mice permit influenza-induced persistent pulmonary dysfunction. Journal of Immunology, 2008. 181(5): p. 3486-94.

[6] Yu, J.S. and N. Bagheri, Multi-class and multi-scale models of complex biological phenomena. Current Opinion in Biotechnology, 2016. 39: p. 167-173.

[7] Hall, M.W., et al., Innate immune function and mortality in critically ill children with influenza: a multicenter study. Critical Care Medicine, 2013. 41(1): p. 224-36.