A Digital Twin of COVID-19 Infected Lungs Allows for Personalized, Interactive Clinical Interventions

This project has received funding from the European Union’s Horizon Europe programme under grant agreement number 101092749. www.crexdata.eu Follow us on social media: A Digital Twin of COVID-19 Infected Lungs Allows for Personalized, Interactive Clinical Interventions Thaleia Ntiniakou1, Charis Akasiadis 2, Hadrien Calmet1, Beatriz Eguzkitza1, Abel Gargallo-Peiró1, Mariano Vázquez1,3, Guillaume Houzeaux1, Alfonso Valencia1,4 and Arnau Montagud1,5 1 Barcelona Supercomputing Center, Barcelona, Spain 2 Institute of Informatics and Telecommunications, NCSR “Demokritos”, Agia Paraskevi, Greece 3 ELEM Biotech SL, Barcelona, Spain 1 CrexData: Critical Action Planning over Extreme-Scale Data ●Combines machine learning & simulations to support Prediction-as-a-Service. ●Uses real-time data streaming and fusion from multiple sources. ●Powers Digital Twin applications—like COVID-19 lung modeling—to support urgent medical decisions. ●Early Time Series Classification (ETSC) enables early detection of non-useful trajectories, saving HPC resources. ● Streaming via Kafka allows real-time monitoring and intervention during simulations. • Build a digital twin of the COVID-19 infected lung to forecast patient-specific outcomes and optimize treatments. • COVID19 treatment digital twin • Simulate lung tissue from cell-level up to the organ-level to build a multiscale, multicellular, spatiotemporal model infected by the SARS-CoV-2 virus. • Infer patient/disease outcomes • Patient-specific predictions to guide treatment decisions. • Drugs • Air ventilator • Tools • PhysiBoSS for the cell level : Agent Based Simulator of multicellular tissues. •Alya for the organ level: air transport arriving to the lungs and the deposition of viral particles in alveoli. The Tools Deposition of the virus particles in the alveoli Multi-physics/multi-scale/ multi-domain simulation tha models: • Air transport arriving to the lungs • The perfusion of the vascular and lymphatic vessels • The transportation of the virus throughout the bronchiole Alya2 Vectorization GPU PhysiBoSS ●Multiscale agent-based simulator ●C++ code adapted to distributed architectures: Models: ●Different cell types of the tissue and their mechanistic models ●The response to O2 ●The diffusion of the virus and how it affects the cells ●The state of the alveolus The coupling of the simulators allow for the digital twin of an infected patient’s treatment Output: which alveoli are dead → another lobe has collapsed Input: Collapsed Alveoli and adjusted flow rate ALYA-LUNG MODEL ALYA-ALVEOLI MODEL PhysiBoSS Output: Quantity of virus for each lung of the 4 lobes Output: Deposition of the viral particles Input: 4 distinct viral deposition patterns DANGER! 40% of the patient's lungs have collapsed! Round 2: Alveolar-Lobe Colapse Input: Collapsed Alveoli and increased flow rate Input: drug intervention → inhibited proteins ALYA-LUNG MODEL ALYA-ALVEOLI MODEL PhysiBoSS Output: Hopefully, no more alveoli are dead Output: Quantity of virus for each lung of the 3 lobes Output: Deposition of the viral particles Round 3: Intervention Abstract The COVID-19 pandemic revealed the urgent need for models bridging scales from molecules to organs to predict outcomes and guide interventions from a mechanistic perspective. Epidemiological and molecular simulations provided insights but remain limited in linking infection dynamics to organ-level pathology and personalized decisions. We present a digital twin of a SARS-CoV-2 infected lung coupling two tools: Alya, which models airflow and viral transport in patient-specific geometries, and PhysiBoSS, an agent-based simulator capturing cell–cell interactions, signaling, and the impact of viral replication and oxygen deprivation in alveoli. A key innovation is a predictive layer atop this modeling. The simulations not only reproduce infection but also forecast trajectories under different scenarios: patient heterogeneity (e.g., vaccination, severity), therapeutic interventions (e.g., drugs, immunomodulation), and supportive care (e.g., ventilation). Iterative cycles provide a systems-level view of disease progression and enable real-time testing of interventions to evaluate organ-level impact. This approach addresses a critical gap: the lack of integrative frameworks combining realistic organ-level transport with mechanistic cell-level models for quantitative, patient-tailored predictions. Our results show the feasibility of this coupling, highlight the role of multiscale interactions in shaping outcomes, and illustrate how digital twins can advance precision medicine. Severe Generic (non personalised) Mild Macrophage BM Epithelial BM p38 FADD NON VACCINATED RECENTLY VACCINATED WANING VACCINATED VIRUS DISTRIBUTION TARGETED DRUGSBOOLEAN MODELSPATIENT TYPES VACCINATION CASP8 downstream MAPK11 =p38 Input: Inhalation Parameters - Flow rate - Particle size (virus) Output: Quantity of virus for each lung of the 5 lobes Output: Deposition of the viral particles 5 distinct viral deposition patterns Output: which alveoli are dead → one lobe has collapsed ALYA-LUNG MODEL ALYA-ALVEOLI MODEL PhysiBoSS Input: 5 distinct viral deposition patterns (user-defined threshold number of infected and dead cells) Round 1: Viral deposition References 1. Ghaffarizadeh A, Heiland R, Friedman SH, Mumenthaler SM, Macklin P. PhysiCell: an open source physics-based cell simulator for 3D multicellular systems. PLoS Comput Biol. 2018;14(2):e1005991. 2. Ponce-de-León M, Ntiniakou T, Montagud A, et al. PhysiBoSS 2.0: a sustainable integration of stochastic Boolean and agent-based modelling frameworks. NPJ Syst Biol Appl. 2023;9:54. 3. Getz M, et al. Modeling SARS-CoV-2 infection in human alveolar tissue. bioRxiv. 2021. 4. Vázquez R, Houzeaux G, et al. Alya: High performance computational mechanics. Arch Comput Methods Eng. 2016;23: 449-478. Contact Thaleia Ntiniakou, [email protected] Conclusions 1. Coupling Alya (organ-level) and PhysiBoSS (cell-level) enables a patient-specific lung infection digital twin. 2. The framework can forecast outcomes under different conditions (vaccination, drug treatments, ventilation). 3. Interactive workflows (Kafka, RapidMiner GUI) allow real-time adaptation of simulations. 4. Early Time Series Classification (ETSC) and HPC orchestration improve scalability and efficiency. Real time Graphical User Interface helps implement the workflow The Crexdata project 4 ICREA, Pg. Lluís Companys 23. Barcelona, Spain 5 Institute for Integrative Systems Biology (I2SysBio), CSIC-UV, Valencia, Spain Ponce-de-León, et al. (2023). Npj Syst Biol App, 9, 54. Getz M. et.al. 2021, bioarchiv