AI discovery of multi-modal immunomodulatory control of sepsis in cases without effective antimicrobials
Author(s):
R. Chase Cockrell; Dale Larie; Gary An
Background:
Despite a great deal of interest in the application of artificial intelligence (AI) to sepsis/critical illness, most current approaches are limited in their potential impact: prediction models do not address the lack of effective therapeutics and many existing optimization approaches cannot address the development of new treatment options. The inability to test new therapeutic applications was highlighted by the generally unsatisfactory results from drug repurposing efforts in COVID19.
Hypothesis:
Addressing this challenge requires the application of model-based deep reinforcement learning (DRL) in a fashion akin to training the game-playing AIs from Deep Mind. We have previously demonstrated the potential of this method in bacterial sepsis where antimicrobial therapies exist. The current work addresses the control problem of multi-modal, adaptive immunomodulation in the circumstance where there is no effective anti-infective therapy (ala in a novel viral pandemic or in the face of resistant microbes).
Methods:
This is a proof-of-concept study that determines the controllability of sepsis without the ability to suppress the infecting pathogen with drugs. We use as a surrogate system a previously validated agent-based model, the Innate Immune Response Agent-based Model (IIRABM), for control discovery using DRL. DRL trains an AI on simulations of infection where the control space is limited to the augmentation or inhibition of immune mediators included in the IIRABM. Policies were learned using gradient descent with the objective function being a return to baseline system health. Generalizability of the discovered control policy was tested on a cohort of alternative parameters/initialconditions with mortalities ranging from 75-85%.
Results:
DRL trained an AI policy that improved system mortality from 85% to 12%. Control actions primarily targeted 3 different aspects of the immune response: 2nd order pro-inflammation governing TH1/TH2 balance, primary anti-inflammation, and inflammatory cell proliferation. Generalizability of the AI policy to systems with different immune responses and pathogen virulence was performed on a set of parameterizations/initial conditions with a mortality rate between 75-85% (N = 1190). The application of the AI (with no additional training or updating) was able to improve survival to 92.7%.
Conclusions:
The current treatment of sepsis is hampered by limitations in therapeutic options able to affect the biology of sepsis. This is heightened in circumstances where no effective antimicrobials exist. Current AI methods are intrinsically unable to address this problem; doing so requires training AIs in contexts that fully represent the counterfactual space of potential treatments. The synthetic data needed for this task is only possible through the use of high-resolution, mechanism-based simulations. Finally being able to treat sepsis will require a reorientation as to the sensing and actuating requirements needed to develop these simulations, train AIs and bring them to the bedside.