Nested active learning for efficient model contextualization and parameterization: pathway to generating simulated populations using multi-scale computational models

A design specification for Critical Illness Digital Twins (CIDTs) to cure sepsis: responding to the National Academies of Sciences, Engineering and Medicine Report "Foundational Research Gaps and Future Directions for Digital Twins"

On December 15, 2023, The National Academies of Sciences, Engineering and Medicine (NASEM) released a report entitled: "Foundational Research Gaps and Future Directions for Digital Twins." The ostensible purpose of this report was to bring some structure to the burgeoning field of digital twins by providing a working definition and a series of research challenges that need to be addressed to allow this technology to fulfill its full potential. In the work presented herein we focus on five specific findings from the NASEM Report: 1) definition of a Digital Twin, 2) using "fit-for-purpose" guidance, 3) developing novel approaches to Verification, Validation and Uncertainty Quantification (VVUQ) of Digital Twins, 4) incorporating control as an explicit purpose for a Digital Twin and 5) using a Digital Twin to guide data collection and sensor development, and describe how these findings are addressed through the design specifications for a Critical Illness Digital Twin (CIDT) aimed at curing sepsis.

The Wound Environment Agent-based Model (WEABM): a digital twin platform for characterization and complex therapeutic discovery for volumetric muscle loss

Volumetric Muscle Loss (VML) injuries are characterized by significant loss of muscle mass, usually due to trauma or surgical resection, often with a residual open wound in clinical settings and subsequent loss of limb function due to the replacement of the lost muscle mass with non-functional scar. Being able to regrow functional muscle in VML injuries is a complex control problem that needs to override robust, evolutionarily conserved healing processes aimed at rapidly closing the defect in lieu of restoration of function. We propose that discovering and implementing this complex control can be accomplished by the development of a Medical Digital Twin of VML. Digital Twins (DTs) are the subject of a recent report from the National Academies of Science, Engineering and Medicine (NASEM), which provides guidance as to the definition, capabilities and research challenges associated with the development and implementation of DTs. Specifically, DTs are defined as dynamic computational models that can be personalized to an individual real world "twin" and are connected to that twin via an ongoing data link. DTs can be used to provide control on the real-world twin that is, by the ongoing data connection, adaptive. We have developed an anatomic scale cell-level agent-based model of VML termed the Wound Environment Agent Based Model (WEABM) that can serve as the computational specification for a DT of VML. Simulations of the WEABM provided fundamental insights into the biology of VML, and we used the WEABM in our previously developed pipeline for simulation-based Deep Reinforcement Learning (DRL) to train an artificial intelligence (AI) to implement a robust generalizable control policy aimed at increasing the healing of VML with functional muscle. The insights into VML obtained include: 1) a competition between fibrosis and myogenesis due to spatial constraints on available edges of intact myofibrils to initiate the myoblast differentiation process, 2) the need to biologically "close" the wound from atmospheric/environmental exposure, which represents an ongoing inflammatory stimulus that promotes fibrosis and 3) that selective, multimodal and adaptive local mediator-level control can shift the trajectory of healing away from a highly evolutionarily beneficial imperative to close the wound via fibrosis. Control discovery with the WEABM identified the following design principles: 1) multimodal adaptive tissue-level mediator control to mitigate pro-inflammation as well as the pro-fibrotic aspects of compensatory anti-inflammation, 2) tissue-level mediator manipulation to promote myogenesis, 3) the use of an engineered extracellular matrix (ECM) to functionally close the wound and 4) the administration of an anti-fibrotic agent focused on the collagen-producing function of fibroblasts and myofibroblasts. The WEABM-trained DRL AI integrates these control modalities and provides design specifications for a potential device that can implement the required wound sensing and intervention delivery capabilities needed. The proposed cyber-physical system integrates the control AI with a physical sense-and-actuate device that meets the tenets of DTs put forth in the NASEM report and can serve as an example schema for the future development of Medical DTs.

Characterizing the Crosstalk Between Programmed Cell Death Pathways in Cytokine Storm With an Agent-Based Model

Background: There is increasing recognition of extensive crosstalk between programmed cell death pathways (PCDPs), such as apoptosis, pyroptosis, and necroptosis, resulting in a highly redundant system responsive to a breadth of potential pathogens. However, because pyroptosis and necroptosis propagate inflammation, these redundancies also present challenges for therapeutic control of dysregulated hyperinflammation seen in cytokine storm (CS) generated organ dysfunction. Hypothesis: We hypothesize that the conversion of existing knowledge regarding apoptosis, pyroptosis, and necroptosis into a computational model can enhance our understanding of the crosstalk between PCDPs via simulation experiments of microbe interactions and experimental interventions. Materials and Methods: Literature regarding apoptosis, pyroptosis, and necroptosis was reviewed and transposed into an agent-based model, the programmed cell death agent-based model (PCDABM). Computational experiments were performed to simulate the activation of various PCDPs as seen by differing microbes, specifically: influenza A virus (IAV), enteropathic Escherichia coli (EPEC), and Salmonella enterica (SE). The potential protective value of PCDP crosstalk was evaluated by silencing either pyroptosis, necroptosis, or both. Computational experiments were also performed simulating the effect of potential therapies blocking tumor necrosis factor (TNF) and interleukin (IL)-1. Results: The PCDABM was implemented in the agent-based modeling toolkit NetLogo. Computational experiments of infection with IAV, EPEC, and SE reproduced cross-activation of PCDPs with effective microbial clearance. Simulations of anti-TNF and anti-IL-1 did not reduce the aggregated amount of inflammation-generated system damage, the surrogate for CS-generated tissue damage. Conclusions: Redundancies have evolved in host PCDPs to maintain protection against a wide range of pathogens. However, these redundancies also challenge attempts at dampening the pathogenic hyperinflammatory state of CS using therapeutic immunomodulation. Integrative simulation models such as the PCDABM can aid in identifying potentially targetable inflection points to mitigate CS while maintaining effective host defense.

Combination Chemotherapy of Multidrug-resistant Early-stage Colon Cancer: Determining Optimal Dose Schedules by High-performance Computer Simulation

The goal of this project was to utilize mechanistic simulation to demonstrate a methodology that could determine drug combination dose schedules and dose intensities that would be most effective in eliminating multidrug resistant cancer cells in early-stage colon cancer. An agent-based model of cell dynamics in human colon crypts was calibrated using measurements of human biopsy specimens. Mutant cancer cells were simulated as cells that were resistant to each of two drugs when the drugs were used separately. The drugs, 5-flurouracil and sulindac, have different mechanisms of action. An artificial neural network was used to generate nearly two hundred thousand two-drug dose schedules. A high-performance computer simulated each dose schedule as a in silico clinical trial and evaluated each dose schedule for its efficiency to cure (eliminate) multidrug resistant cancer cells and its toxicity to the host, as indicated by continued crypt function. Among the dose schedules that were generated, 2430 dose schedules were found to cure all multidrug resistant mutants in each of the 50 simulated trials and retained colon crypt function. One dose schedule was optimal; it eliminated multidrug resistant cancer cells with the minimum toxicity and had a time schedule that would be practical for implementation in the clinic. These results demonstrate a procedure to identify which combination drug dose schedules could be most effective in eliminating drug resistant cancer cells. This was accomplished using a calibrated agent-based model of a human tissue, and a high-performance computer simulation of clinical trials.

Preparing for the next pandemic: Simulation-based deep reinforcement learning to discover and test multimodal control of systemic inflammation using repurposed immunomodulatory agents

Background

Preparation to address the critical gap in a future pandemic between non-pharmacological measures and the deployment of new drugs/vaccines requires addressing two factors: 1) finding virus/pathogen-agnostic pathophysiological targets to mitigate disease severity and 2) finding a more rational approach to repurposing existing drugs. It is increasingly recognized that acute viral disease severity is heavily driven by the immune response to the infection ("cytokine storm" or "cytokine release syndrome"). There exist numerous clinically available biologics that suppress various pro-inflammatory cytokines/mediators, but it is extremely difficult to identify clinically effective treatment regimens with these agents. We propose that this is a complex control problem that resists standard methods of developing treatment regimens and accomplishing this goal requires the application of simulation-based, model-free deep reinforcement learning (DRL) in a fashion akin to training successful game-playing artificial intelligences (AIs). This proof-of-concept study determines if simulated sepsis (e.g. infection-driven cytokine storm) can be controlled in the absence of effective antimicrobial agents by targeting cytokines for which FDA-approved biologics currently exist.

Methods

We use a previously validated agent-based model, the Innate Immune Response Agent-based Model (IIRABM), for control discovery using DRL. DRL training used a Deep Deterministic Policy Gradient (DDPG) approach with a clinically plausible control interval of 6 hours with manipulation of six cytokines for which there are existing drugs: Tumor Necrosis Factor (TNF), Interleukin-1 (IL-1), Interleukin-4 (IL-4), Interleukin-8 (IL-8), Interleukin-12 (IL-12) and Interferon-γ(IFNg).

Results

DRL trained an AI policy that could improve outcomes from a baseline Recovered Rate of 61% to one with a Recovered Rate of 90% over ~21 days simulated time. This DRL policy was then tested on four different parameterizations not seen in training representing a range of host and microbe characteristics, demonstrating a range of improvement in Recovered Rate by +33% to +56.

Discussion

The current proof-of-concept study demonstrates that significant disease severity mitigation can potentially be accomplished with existing anti-mediator drugs, but only through a multi-modal, adaptive treatment policy requiring implementation with an AI. While the actual clinical implementation of this approach is a projection for the future, the current goal of this work is to inspire the development of a research ecosystem that marries what is needed to improve the simulation models with the development of the sensing/assay technologies to collect the data needed to iteratively refine those models.

Drug Development Digital Twins for Drug Discovery, Testing and Repurposing: A Schema for Requirements and Development

There has been a great deal of interest in the concept, development and implementation of medical digital twins. This interest has led to wide ranging perceptions of what constitutes a medical digital twin. This Perspectives article will provide 1) a description of fundamental features of industrial digital twins, the source of the digital twin concept, 2) aspects of biology that challenge the implementation of medical digital twins, 3) a schematic program of how a specific medical digital twin project could be defined, and 4) an example description within that schematic program for a specific type of medical digital twin intended for drug discovery, testing and repurposing, the Drug Development Digital Twin (DDDT).

Preparing for the next COVID: Deep Reinforcement Learning trained Artificial Intelligence discovery of multi-modal immunomodulatory control of systemic inflammation in the absence of effective anti-microbials

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 (and cannot) address the lack of effective therapeutics and current approaches to enhancing the treatment of sepsis focus on optimizing the application of existing interventions, and thus cannot address the development of new treatment options/modalities. The inability to test new therapeutic applications was highlighted by the generally unsatisfactory results from drug repurposing efforts in COVID-19.

Hypothesis

Addressing this challenge requires the application of simulation-based, model-free deep reinforcement learning (DRL) in a fashion akin to training the game-playing AIs. We have previously demonstrated the potential of this method in the context of bacterial sepsis in which the microbial infection is responsive to antibiotic therapy. The current work addresses the control problem of multi-modal, adaptive immunomodulation in the circumstance where there is no effective anti-pathogen therapy (e.g., 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 pharmacologically suppress the pathogen. 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. The DRL algorithm 'trains' an AI on simulations of infection where both the control and observation spaces are limited to operating upon the defined immune mediators included in the IIRABM (a total of 11). Policies were learned using the Deep Deterministic Policy Gradient approach, with the objective function being a return to baseline system health.

Results

DRL trained an AI policy that improved system mortality from 85% to 10.4%. Control actions affected every one of the 11 targetable cytokines and could be divided into those with static/unchanging controls and those with variable/adaptive controls. Adaptive controls primarily targeted 3 different aspects of the immune response: 2nd order pro-inflammation governing TH1/TH2 balance, primary anti-inflammation, and inflammatory cell proliferation.

Discussion

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, as was the case for COVID-19. 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 and bring them to the bedside.

The Use of Artificial Neural Networks to Forecast the Behavior of Agent-Based Models of Pathophysiology: An Example Utilizing an Agent-Based Model of Sepsis

Introduction: Disease states are being characterized at finer and finer levels of resolution via biomarker or gene expression profiles, while at the same time. Machine learning (ML) is increasingly used to analyze and potentially classify or predict the behavior of biological systems based on such characterization. As ML applications are extremely data-intensive, given the relative sparsity of biomedical data sets ML training of artificial neural networks (ANNs) often require the use of synthetic training data. Agent-based models (ABMs) that incorporate known biological mechanisms and their associated stochastic properties are a potential means of generating synthetic data. Herein we present an example of ML used to train an artificial neural network (ANN) as a surrogate system used to predict the time evolution of an ABM focusing on the clinical condition of sepsis. Methods: The disease trajectories for clinical sepsis, in terms of temporal cytokine and phenotypic dynamics, can be interpreted as a random dynamical system. The Innate Immune Response Agent-based Model (IIRABM) is a well-established model that utilizes known cellular and molecular rules to simulate disease trajectories corresponding to clinical sepsis. We have utilized two distinct neural network architectures, Long Short-Term Memory and Multi-Layer Perceptron, to take a time sequence of five measurements of eleven IIRABM simulated serum cytokine concentrations as input and to return both the future cytokine trajectories as well as an aggregate metric representing the patient's state of health. Results: The ANNs predicted model trajectories with the expected amount of error, due to stochasticity in the simulation, and recognizing that the mapping from a specific cytokine profile to a state-of-health is not unique. The Multi-Layer Perceptron neural network, generated predictions with a more accurate forecasted trajectory cone. Discussion: This work serves as a proof-of-concept for the use of ANNs to predict disease progression in sepsis as represented by an ABM. The findings demonstrate that multicellular systems with intrinsic stochasticity can be approximated with an ANN, but that forecasting a specific trajectory of the system requires sequential updating of the system state to provide a rolling forecast horizon.