The rise of AI as a powerful computational tool to integrate large datasets presents a special opportunity to deal with the inherent complexity of biological systems. However, machine learning approaches do not consider the mechanistic knowledge of the underlying interactions. Preliminary studies that combine ODEs and machine learning highlight that these computational algorithms could be on the cusp of a major revolution. However, no parameter estimation theory exists to integrate simultaneously both approaches.

We propose to create new hybrid models and test their predictions in a mouse viral coinfection system to address a central vexation for infection biology: how and when to modulate immune responses to mitigate mortality during lethal respiratory viral infection.

To validate and test our novel and foundational mathematical approaches, we will generate the biological data from a mouse infection system with a mild viral pathogen (rhinovirus) two days before infection with a lethal viral pathogen (influenza) that results in reduced disease compared to influenza infection alone. We hypothesize that this system can train our mathematical models in a natural way how the innate immune system can be manipulated to reduce mortality to lethal influenza.

Figure: Hybrid Model Diagram. Mechanistic model predictions are in blue, and Gray-model predictions are in orange. Data integration is shown in green. Arrows show the feedback between the models or order for data integration.