EvoluPark

Parkinson’s disease (PD) is a chronic, neurodegenerative, and heterogeneous condition, involving both severe motor and non-motor manifestations. Despite advances in symptom management, there is currently no tool capable of modeling the future trajectory of the disease for individual patients or predicting the potential effects of specific interventions. Yet, each disease course is shaped by complex interactions between biomarkers, behaviors, and treatments. The EvoluPark project aims to address this gap by developing a multiscale Medical Digital Twin (MDT) capable of integrating and leveraging clinical, biological, structural, behavioral, and patient-reported data. This MDT will thus integrate five key data types: Clinical – motor and cognitive scores ; Molecular

– detection of α-synuclein aggregates by RT-QuIC (Real-Time Quaking-Induced Conversion);

Retinal – imaging via OCT and fundus photography ; Behavioral – analysis of gait, speech, and handwriting ; Subjective – self-reported Quality of Life (QoL) questionnaires.

These data will feed into an application-agnostic AI architecture organized around five key modules, specifically designed for data-scarce environments:

1) Perception: extraction of multimodal latent representations using foundation models tailored to each data type.

2) Temporal forecasting: self-supervised trajectory learning using Joint Embedding Predictive Architectures (JEPA), robust to irregular and missing data.

3) Simulation: stochastic projection of future states using latent diffusion models, with uncertainty estimation and intervention simulation.

4) Personalization: individual-level adaptation using LoRA (Low-Rank Adaptation) modules, ensuring both precision and generalizability.

5) Stratification: discovery of clinically meaningful subgroups (e.g., dopamine-responsive vs. resistant profiles) from simulated trajectories to support clinical decision-making.

Special attention is given to harmonizing data streams with different sampling frequencies: QoL data, collected monthly, and other assessments, collected semiannually, are handled by a dynamic temporal encoder capable of processing asynchronous inputs effectively.

The consortium brings together complementary expertise in AI (UBO–LaTIM), molecular biology (INRAe–IHAP & UBO-LaTIM), neurology and ophthalmology (CHU Brest & Orléans), and patient engagement (CECAP). Data collection will be coordinated via the Inserm CIC 1412 in Brest, with secure hosting through HDS-certified provider Eskemm. ADCIS company will support interface development for patients and clinicians, and SATT Ouest Valorisation will manage intellectual property and valorization.

Expected outcomes include: a clinically validated MDT prototype for PD monitoring and simulation ; adaptive digital twins that evolve with individual data ; clinically validated stratification ; open-access publications, software, and datasets.

In alignment with the PEPR Digital Health initiative, EvoluPark combines algorithmic innovation, clinical robustness, and patient-centered design. It inaugurates a new generation of digital twins—evolving with and for the patient—linking molecular, behavioral, subjective, and functional levels in an explainable, modular, and clinically actionable framework. In doing so, EvoluPark marks a step toward the digital continuum of medicine and a new era of personalized, proactive care.