Learning tumour language: Neovivum’s role in predicting cancer resistance

Current precision oncology faces a fundamental limitation. While doctors can identify biomarkers that correlate with treatment response, they cannot model how tumours adapt when therapy applies pressure. A patient receives a drug, the tumour initially shrinks, but within months resistant cells emerge, and clinicians are left reacting to resistance rather than anticipating it.

Tumours are not static masses but dynamic ecosystems where cancer cells, immune infiltrates, blood vessels, and stromal components communicate constantly. This cellular crosstalk determines whether treatments succeed or fail. Neovivum Technologies, as technical lead in the CancerScan project, is building the first computational framework to learn the transformation rules governing these communication networks – what the consortium calls tumour “proto-grammar.”

“Tumours don’t just resist treatment randomly. They follow rules, like a language with its own grammar. If we can learn this ‘proto-grammar’ of tumour communication, we believe we can predict how they’ll adapt before resistance emerges. That’s the shift from reactive to anticipatory oncology,” explains Igor Balaž, Project Coordinator at Neovivum Technologies.

Orchestrating multiple technologies into a unified framework

CancerScan brings together seven European partners, each contributing specialized expertise. Politecnico di Milano develops GPU-FPGA computing infrastructure to accelerate simulations. University of Bielefeld reconstructs cellular communication networks from single-cell data. FCiências.ID builds knowledge graphs integrating tumour biology databases. Vall d’Hebron Hospital provides organoid experiments and clinical validation data.

Neovivum’s role is to orchestrate these streams into a cohesive digital twin framework. This involves three critical technical challenges. First, automated parameter extraction converts pathology slides and molecular data into simulation-ready inputs. Second, event-driven knowledge graph integration ensures the digital twin adjusts parameters dynamically as virtual tumour composition changes during simulated treatment. Third, ensemble modelling quantifies prediction uncertainty by running multiple simulation scenarios with varying assumptions.

The validation standard is rigorous. Digital twin predictions must achieve at least 95% agreement with experimental organoid data and demonstrate 20% improvement over existing computational methods when tested against patient biopsy outcomes. To meet clinical deployment constraints, the framework must reduce simulation time from weeks to minutes, a challenge addressed through close collaboration with Politecnico di Milano’s hardware acceleration work.

From concept to clinical tool

Neovivum originated the CancerScan concept by recognizing that tumour microenvironment communication follows learnable patterns. While traditional quantitative systems pharmacology models treat tumours as homogeneous targets responding predictably to drugs, reality is messier. Tumour cells negotiate with their environment, recruiting blood vessels, suppressing immune responses, and modulating stromal barriers, all through molecular signalling that can be mapped, analysed, and ultimately predicted.

The proto-grammar framework uses two complementary approaches. Statistical pattern analysis identifies interaction coefficients between cell types across different tumour compositions. Deep learning attention networks, similar to architectures powering large language models, learn how communication networks transform under treatment and microenvironment context. By combining these methods, the framework captures both mechanistic biological insights and data-driven prediction power.

Currently in Month 6 of the 36-month project, Neovivum has delivered the comprehensive test plan outlining validation methodology and is coordinating data specification agreements with experimental partners. The next critical milestone is establishing parameter mapping protocols that translate qualitative network analysis into quantitative simulation coefficients, scheduled for resolution through joint workshops in the coming months.

Beyond CancerScan: building the foundation for evolutionary medical AI

Neovivum’s broader vision extends beyond pancreatic cancer digital twins. The company’s research philosophy centres on evolutionary intelligence, AI systems that adapt and self-optimize through iterative learning, mirroring biological evolution. The modular digital twin framework developed in CancerScan represents a foundational architecture applicable to other cancer types and potentially other diseases where cellular communication drives pathology.

The commercialization pathway encompasses two complementary products: the integrated clinical platform combining smart pathology scanning, automated analysis, and digital twin simulation for hospital deployment, and the modular digital twin framework as standalone technology for pharmaceutical research and academic oncology. Neovivum leads commercialization strategy for both tracks, enabling the consortium to pursue clinical deployment collaboratively while accelerating impact across research and drug development sectors independently.

By learning the language tumours speak, CancerScan aims to transform oncology from reactive treatment selection to anticipatory, personalised intervention. Neovivum Technologies is building the computational foundation that makes this vision possible.

Links

Neovivum website: https://neovivum.com

Neovivum LinkedIn: https://www.linkedin.com/company/neovivum

Neovivum X: https://x.com/neo_vivum

CancerScan project website: https://www.cancerscanproject.eu/

LinkedIn channel: http://www.linkedin.com/company/cancerscan-project

X channel: https://x.com/CancerScan_eu

YouTube channel: https://www.youtube.com/@cancerscan-Project

Keywords

tumour proto-grammar, digital twin, precision oncology, cellular communication networks, evolutionary intelligence, parameter extraction, ensemble modelling, knowledge graph integration, treatment resistance prediction, personalized cancer therapy