Graph representation forecasting of patient's medical conditions: towards a digital twin

Pietro Barbiero, Ramón Viñas Torné, Pietro LióarXiv 2020 · arXiv

Abstract: This paper introduces a modular AI framework designed to create "healthcare digital twins"—virtual models that mirror a patient's unique physiological state. By integrating Graph Neural Networks (GNNs) for organ-level interaction modeling and Generative Adversarial Networks (GANs) for high-dimensional molecular data synthesis, the framework bridges the gap between mechanistic biological models and deep learning. The system demonstrates the ability to forecast disease trajectories, such as hypertension and viral infections, and simulate counterfactual treatment scenarios. Results indicate that this hybrid approach can provide personalized, probabilistic predictions of health outcomes, marking a significant step toward preventative precision medicine.

1. Introduction

Modern medicine is predominantly reactive, addressing symptoms only after they manifest. Precision medicine seeks to shift this paradigm toward personalized prevention. However, the human body is a multiscale system where molecular changes in one tissue can trigger systemic "ripple effects" across organ systems. While mechanistic models, such as systems of ordinary differential equations (ODEs), offer high interpretability, they often struggle to integrate the massive, heterogeneous datasets generated by modern diagnostics. Conversely, standard deep learning models excel at data integration but frequently lack biological grounding. This work proposes a hybrid "digital twin" architecture to bridge these approaches.

Conceptual overview of the digital twin framework, illustrating the mapping from multi-scale biological data to healthy and diseased trajectories.

2. Methodology

The proposed architecture is modular, allowing for the integration of disparate data types and established biological knowledge.

2.1 Graph Neural Networks (GNNs)

The patient is represented as a graph where nodes correspond to biological entities (e.g., organs, tissues, or processes) and edges represent physiological signals or relationships. Message-passing mechanisms within the GNN simulate how localized changes, such as lung inflammation, propagate to affect other systems like heart rate or blood pressure.

2.2 Generative Adversarial Networks (GANs)

To address the scarcity of "paired" datasets—where every layer of a patient's biology is measured simultaneously—the framework employs a Wasserstein GAN with Gradient Penalty (WGAN-GP). This component generates synthetic transcriptomic (gene expression) data, allowing the system to impute missing biological information or simulate rare clinical scenarios.

2.3 Mechanistic Integration

The system incorporates handcrafted ODE models to define the initial graph topology and generate baseline training data. This ensures that the AI's predictions remain consistent with known laws of physiology, creating a "gray-box" model that combines the strengths of both mechanistic and data-driven approaches.

The comprehensive modular architecture of the digital twin, detailing the interaction between the GAN-based generator, the message-passing GNN, and the underlying ODE systems.

3. Experimental Setup and Datasets

The framework was validated using several high-dimensional biological resources:

  • GTEx (Genotype-Tissue Expression) Project: Used to train the GAN on cross-tissue gene expression patterns.
  • KEGG Pathway Database: Utilized to identify specific gene networks involved in the Renin-Angiotensin System (RAS) and inflammatory pathways.
  • Synthetic ODE Data: Time-series data from established cardiovascular models were used to train the GNN's forecasting capabilities.

4. Key Results and Discussion

The researchers demonstrated several critical capabilities of the digital twin framework:

  • Transcriptomic Crosstalk: The model successfully predicted tissue-specific gene expression from blood-borne signaling factors, validating the message-passing approach.
  • Cardiovascular Forecasting: In simulations of elderly patients with hypertension and viral infections (e.g., SARS-CoV-2), the GNN accurately forecasted blood pressure trajectories over time.
  • Virtual Clinical Trials: The system simulated the impact of treatments like Benazepril and Heparin, showing reductions in both mean blood pressure and its variability—a key clinical risk factor.
  • Uncertainty Quantification: By employing Monte Carlo Dropout, the model provides 95% confidence intervals, offering a probabilistic view of a patient's future rather than a rigid, deterministic prediction.

5. Conclusion and Future Work

This modular framework offers a scalable path toward holistic digital twins in healthcare. While currently limited by the availability of real-world paired datasets, the system's ability to simulate counterfactuals and quantify uncertainty provides a powerful tool for personalized clinical decision support. Future work will focus on expanding the model to include environmental factors (the "exposome") and enhancing the transparency of the neural network components for clinical trust.

Summary updated

Abstract

Objective: Modern medicine needs to shift from a wait and react, curative discipline to a preventative, interdisciplinary science aiming at providing personalised, systemic and precise treatment plans to patients. The aim of this work is to present how the integration of machine learning approaches with mechanistic computational modelling could yield a reliable infrastructure to run probabilistic simulations where the entire organism is considered as a whole. Methods: We propose a general framework that composes advanced AI approaches and integrates mathematical modelling in order to provide a panoramic view over current and future physiological conditions. The proposed architecture is based on a graph neural network (GNNs) forecasting clinically relevant endpoints (such as blood pressure) and a generative adversarial network (GANs) providing a proof of concept of transcriptomic integrability. Results: We show the results of the investigation of pathological effects of overexpression of ACE2 across different signalling pathways in multiple tissues on cardiovascular functions. We provide a proof of concept of integrating a large set of composable clinical models using molecular data to drive local and global clinical parameters and derive future trajectories representing the evolution of the physiological state of the patient. Significance: We argue that the graph representation of a computational patient has potential to solve important technological challenges in integrating multiscale computational modelling with AI. We believe that this work represents a step forward towards a healthcare digital twin.
BibTeX
@misc{barbiero2020graph,
  author = {Pietro Barbiero and Ramón Viñas Torné and Pietro Lió},
  title = {{Graph representation forecasting of patient's medical conditions: towards a digital twin}},
  year = {2020},
  eprint = {2009.08299},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2009.08299},
}
Figure