Methodology

Biodigital Twin Methodology

The realization of the ADT follows a rigorous methodology to realize three different objectives on the course to a full functional digital twin.

Methodology I

Dynamic Equilibrium

Systems Physiology Models represent multi-scale feedback loops and internal states of cells, tissues, organs…all the way to organism functional performance…that maintain a dynamic equilibrium in response to nutritional, perfusion, utilization, and other perturbations (including utilization in normal or µ gravity).  These models are formulated mathematically as differential equations, a critical representation for capturing homeostatic stability, and the shift in dynamic equilibrium in response to perturbations and population variants.

Methodology II

Digital Population

The behavior of the model is rigorously tested against a scientific/medical advisory board-approved clinical validation data (both terrestrial and flight data).  Population variants in model parameters create a high-dimensional hypothesis space; population data constrains the feasibility of this hypothesis space, and Bayesian inference is used to determine the relative likelihood of each hypothesis such that the resulting Digital Population reproduces all population statistics chosen for validation. Bayesian inference is a statistical inference method that uses Bayes’ theorem to revise the probability of a hypothesis as new evidence or information is available.

Methodology III

Digital Twin

The probability-weighted hypotheses of the Digital Population serve as a Bayesian prior for generating an astronaut’s Digital Twin using the astronaut’s data for Bayesian inference.  The Digital Twin continues to learn and improve with the latest Twin being used as a prior for successive rounds of Bayesian inference.  In a manner parallel to multiple hurricane models providing probabilistic forecasts during hurricane season, the latest Twin automatically generates a set of “What-If?” simulation results that are provided for alerts or updated countermeasure plans.

    ADT Application

    There are many applications of a fully realized ADT Platform, which are ultimately data- and context-dependent. Beyond the summary of clinical features, the ADT supports the following application activities: 

    Data Sources

    Much of the data that feeds into the foundational development of the Astronaut Digital Twin is of terrestrial origin. In addition to the wealth of Earth-based data, the ADT also utilizes historical data from more than 60 years of humans in space, including data from recent missions that are rich in multiomic features. These multiomic* studies include: NASA Twins Study of One Year in Space (341 days in space), SpaceX Inspiration4 (3-days in space), Axiom-2 (12 days in space), and Axiom-3 (21 days in space). We are also involved in several missions in the development stage.

    Because our team is involved in the original research at the pre-analytical, analytical, data curation, and interpretation phase of the multiomics experiments in space, we have additional insight into how such data may be optimized for inclusion in development of the astronaut digital twin.

    *Multiomics is a molecular analysis approach in which the analytes are not preselected (i.e., untargeted), though targeted analyses are often conducted. Included in a multiomics analysis are commonly the genome, epigenome, transcriptome, proteome, metabolome, microbiome, and exposome (or a subset thereof).  

    Digital Twins and the AI Landscape

    The Astronaut Digital Twin, AI, and machine learning are complementary elements that align to the specific objective of designing a solution where the digital twin represents the physical human (or its subsystems) as closely as possible. There are several elements that support this view as to the necessity and benefit of the Astronaut Digital Twin:

    1. Fragmented studies: most studies are fragmented and usually lack the common parameters necessary for AI/ML to learn effectively 
    2. Lack of observations and features: artificial Intelligence and machine learning require large datasets. Biological datasets, especially those that have been gathered from the spaceflight environment with small astronaut cohorts, are not “big data” in the classical sense 
    3. Lack of fundamental, causative framework: with no engine to animate and interpret data, findings are mostly correlative, lacking causative reasoning and understanding

    The Astronaut Digital Twin platform provides an advancement in attempting to solve these issues with its ability to aggregate fragmented studies/data and animate those studies quantitatively in a causative framework and a human-in-the-loop systems engineering-based, iterative construction. 

    Artificial intelligence and machine learning methodologies are valuable to improving the efficiency of the Astronaut Digital Twin in the following ways: 

    • Optimization (twin creation)
    • Optimization (twin updating)
    • Generative modeling
    • Data analytics
    • Predictive analytics
    • Decision making and support