Critical Illness Digital Twins:

Insight from the Fusion of Data-driven and Mechanistic Modeling

Yoram Vodovotz, Ph.D.

​

Departments of Surgery, Immunology, Computational & Systems Biology, Clinical and Translational Science, and Bioengineering

McGowan Institute for Regenerative Medicine

Center for Systems Immunology

University of Pittsburgh

​

​

​

​

​

​

2024 GLIMPRINT/IMAG-MSM Seminar

Disclosures, Funding, �and Acknowledgements

• National Institutes of Health
• Department of Defense/DARPA

​

• Co-founder and stakeholder in Immunetrics, Inc.
• Chief Scientific Advisor, Anuna AI, Inc.
• Field Chief Editor, Frontiers in Systems Biology

Disclosures

Funding

Inflammation is…

• The body’s way of informing itself of changes in homeostasis, either from without or within
• Necessary for proper healing and regeneration
• A key component of the response to stress
• Intertwined with other responses (e.g., coagulation)
• Complex, redundant, interconnected
• Nonlinear trajectories
• Feedbacks and thresholds
• Manifests as dynamic patterns
• Dysregulated in many diseases 🡪 critical illness focus

​

• A puzzle: inflammation can be both good and bad

​

Pediatric

Acute Liver

Failure

Trauma

Sepsis/COVID-19

Wounds

Trauma

Healthy vs. Unhealthy Aging

Inflammation Across life- & health-span

Unifying data across time and space via data-driven and mechanistic modeling

​

Controlling inflammation using AI

It’s a mindset assisted by a toolset

Digital Twins: �Integrating data-driven & mechanistic models�

Data-Driven Models

Association-based, may suggest principal drivers & dynamic networks

Limited use of prior knowledge

Require large amounts of data

Prediction within “training data” …yet often used incorrectly to predict outside such data

Mechanistic Models

Highly causal, may suggest non-intuitive aspects

Extensive use of prior knowledge

Allow explanation of emergent phenomena

More difficult

Predictions outside of “training data”

Digital Twins:�An Emerging Paradigm for Complex Systems

Inflammation “Digital Twin” 1.0:�Local inflammation, lessons learned

• Mechanistic, agent-based model (ABM) used to simulate phonation injury to the vocal folds in humans
• ABM simulations reproduced trajectories of inflammatory mediators in laryngeal secretions up to 4h post-injury
• ABM predicted levels of these mediators at 24h post-injury
• In silico trials of spontaneous speech vs. rest vs. resonant voice exercise

Could we do the same for critical illness?

Critical Illness “Digital Twin” 1.0:�Systemic inflammation

• Blood
• Serum
• Plasma
• Cells
• Metabolites
• Inflammatory Mediators

Con: No insight into organ-specific impact

Major Pro: ability to predict % non-survivors having been trained solely on data from survivors

Pros:

• Critical illness 🡪 systemic inflammation
• Systemic circulation is easily accessible
• Genetic predisposition 🡪 parameter values

Critical Illness “Digital Twin” 2.0:�Cross-compartmental inflammation

Critical Illness “Digital Twin” 2.0:�Cross-compartmental inflammation

Inflammation “Digital Twin” 2.0:�Data-driven Modeling Toolset

Voit et al. Frontiers Systems Biol.. 2023

Chemokine-Related Roles for the Vagus Nerve in Modulating Intra- and Inter-Tissue Inflammation Thresholds Inferred from Experimental and Computational Studies

Can we learn anything from combined intra- and inter-tissue modeling using DyNA and Dynamic Hypergraphs?

Vagotomy disrupts anti-inflammatory inter-tissue connectivity across the

spleen-plasma, kidney-gut, and lung-gut axes

DyNA

DyHyp

Does the vagus nerve serve as

a cross-tissue inflammation threshold?

• ANOVA, DyNA, and DyHyp: vagotomy differentially affects either the steady-state levels and/or rate of change of IFNγ, IP-10, MIG, MCP-1, and IL-6 across one or more tissues
• IP-10/CXCL10: Unique edge in the systemic circulation network following vagotomy but not sham surgery
• IL-6 was not a significant inter- or intra-inflammatory edge in multiple tissues but was elevated significantly following vagotomy relative to sham surgery in the plasma, kidney, and lung
• Hypothesis: Dynamic spatiotemporal expression of IFN-γ and subsequent expression of the IFN-γ-induced chemokines IP-10, MIG, and MCP-1 may influence the rate of change of IL-6 and account for its unique spatiotemporally sensitive expression

Rate-of-change modeling suggests a novel, inter-tissue, vagally regulated inflammatory threshold involving the spleen and impacting IFNγ, IP-10, MIG, MCP-1, and IL-6

LPS disrupts vagally mediated cross-tissue threshold involving IP-10

LPS

Integration of machine learning with mechanistic modeling in trauma patients

“Chemokine Switch”

Encode as a

quasi-Boolean model

• Reproduce dynamic trajectories
• Predict missing nodes
• IP-10 exhibits sigmoidal behavior

Chemokine Switch: Threshold behavior

of IP-10

Zaaqoq et al, Crit. Care Med 2014

Hemorrhage in silico: Hemorrhagic Shock Digital Twins Integrate Large-Animal and Human Inflammation, Coagulation, and Resuscitation Data

Hemorrhage in silico: Toward Hemorrhagic Shock Digital Twins Integrating Large-Animal and Human Inflammation, Coagulation, and Resuscitation Data

Cannon, J.W. et al. Commun. Med. 2024 (2024. 4:113. https://doi.org/10.1038/s43856-024-00535-6)

• Accurate prediction of physiologic, inflammatory, and laboratory measures in both swine and patients
• Prediction of outcome and time of death in the PROMMTT cohort
• Resuscitation with plasma and red blood cells together outperforms resuscitation with crystalloid or plasma alone
• Earlier plasma resuscitation reduced both morbidity and mortality

Workflow may serve as a translational bridge from pre-clinical to clinical studies in T/HS and other complex disease settings

Pattern 🡪 mechanism 🡪 In Silico Trial�

Model-based Precision Medicine

Clinical Validation

“Digital Twins”/

In Silico Clinical Trials

Actuate Control

• An et al. Int. J. Burns Trauma 2012
• Translational Systems Biology: Concepts and Practice for the Future of Biomedical Research. An, G. and Vodovotz, Y. 2014
• Day et al. Curr. Opin. Syst. Biol. 2018
• Vodovotz, Trends in Immunol. 2023

Rational Reprogramming

An Interdisciplinary Team Effort

Surgery (Pitt)

Tim Billiar

Ruben Zamora

Jason Sperry

Andrew Peitzman

Jennifer Steel

Andrew Abboud

Ashti Shah

Lukas Schimunek

Andres Torres

Rajaie Namas

Rami Namas

Khalid Almahmoud

Othman Malak

Andrew Abboud

Fayten el-Dehaibi

Derek Barclay

Jinling Yin

​

McGowan Institute (Pitt)

Steve Badylak

Bryan Brown

Jörg Gerlach

Ira Fox

Alejandro Soto-Gutierrez

​

Center for Systems Immunology (Pitt)

Harinder Singh

Jishnu Das

​

Critical Care Medicine (Pitt)

Derek Angus

Gilles Clermont

Chris Seymour

Mitchell Fink

John Kellum

Murat Kaynar

Akram Zaaqoq

​

​

SHRS (Pitt)

Qi Mi

​

Statistics (Pitt/Pitt Bradford)

Greg Constantine

Marius Buliga

​

Anesthesiology (Pitt)

Yan Xu

Maria Cohen

​

Carnegie Mellon University

Tzahi Cohen-Karni

Doug Weber

​

Indiana University

Todd McKinley

​

Immunetrics / Simulations Plus

John Bartels

Steve Chang

​

Walter Reed National Military Medical Center

Eric Elster

Jonathan Forsberg

Seth Schobel

​

University of Vermont

Gary An

Chase Cockrell

​

Caltech

John Doyle

​

Rutgers University

Ioannis Androulakis

​

​

Univ. of Tennessee—Knoxville / Applied Biomath

Judy Day

​

​

SPECS-lab, Inst. for Bioengineering of Catalunya / Radboud Univ.

Paul Verschure

Riccardo Zucca

Xerxes Asiwalla

​

​

​

​

Email: vodovotz@pitt.edu

CREDITS: This presentation template was created by Slidesgo, and includes icons by Flaticon and infographics & images by Freepik