Context
Healthcare digital twins remain artisanal, specific, costly and reserved for large actors with big data volumes.
Approach
Design of a universal digital twin generator: a platform enabling domain experts to create their own digital twins without algorithmic prerequisites, from compact datasets.
The problem
Healthcare digital twins are currently artisanal: each twin is designed for a specific use, requires significant data volumes, dedicated data science teams, and heavy investments. The result is often a narrow-use model with limited scope and unpredictable quality. Only large players can afford this.
The TweenMe approach
TweenMe is a universal generator of digital twins: from a single engine, the platform can process as many datasets as needed to create as many digital twins as desired, across varied domains (epidemiology, medical research, pharmaceutical R&D, economic modelling).
The architecture relies on a network of reconfigurable specialised AIs rather than a single large model. The pipeline comprises five modules: the Ontology Modeler (data asset qualification and structuring), the Smart Data Fertilizer (dimensional enrichment and volumetric augmentation via generative AI limiting statistical hallucinations), the Bakery (packaging the model into an application usable by the domain expert), the DT Show Room (using produced models in inference via an Open WebUI or FastAPI interface), and the Cockpit (command centre for the various pipelines under execution).
Results
The OCTOPUS study (mNSCLC BRAF V600E, N=184 cohort) demonstrates the pipeline’s capabilities on an HDLSS (High Dimension, Low Sample Size) problem. Survival modelling via SurvTRACE (attention-based transformer) and TSTR (Train on Synthetic, Test on Real) validation achieve 95.2% ML Utility. An integrated counterfactual simulator enables exploration of alternative therapeutic scenarios. This work was selected in the top 5% of ISPOR Europe 2025 posters.
Learnings
TweenMe demonstrates that industrialising healthcare digital twin production is possible with a fundamentally different approach: working with compact data, qualifying before modelling, and packaging the result in an interface accessible to the domain expert unfamiliar with AI. The downscale approach (little data, thorough qualification) opposes the dominant upscale paradigm (lots of data, minimal qualification).
Videos and presentations
Introduction to TweenMe
AFCRO 2024 Pitch — Clinical Research Innovation Forum
Elevator pitch at the AFCROs Forum, January 25, 2024. First public presentation of the universal digital twin generator concept.
▶ Watch on YouTube → · Read the publication →
Qualees Corporate Presentation: AFCRO 2025, March 11, 2025
After presenting the concept in 2024, we present TweenMe as an operational universal Do It Yourself digital twin generator.
▶ Watch on YouTube → · Read the publication →