PharosAI partners with 10x Genomics on spatial cancer data backed by £18.9 million

The initiative is backed by £18.9 million (€21.8 million) from the U.K. government’s Research Ventures Catalyst, with additional partner support, and will start with breast cancer before expanding to lung and pancreatic cancer, with the program running through 2027.

The collaboration fits into a growing push to make cancer data more usable for computational research. PharosAI’s stated aim is not only to digitise tumor tissue into structured datasets, but to pair them with AI models and analytical tools and to make the resulting resources available to a wider research and innovation community.

Inside the PharosAI-10x Genomics collaboration

In this collaboration, the consortium will work primarily with existing tumor specimens stored by NHS hospitals, converting physical tissue into structured digital datasets that can be analyzed at scale across patient cohorts.

The analytical backbone of that effort is 10x Genomics’ Xenium platform, which allows researchers to measure gene expression directly within intact tissue sections, rather than after samples are dissociated. This means tumour samples are processed on-slide, preserving the organization of cancer cells, immune cells, and surrounding tissue while capturing molecular readouts across the tissue section.

For 10x Genomics, the collaboration represents a high-profile deployment of its platform. The California-based company is best known for its single-cell and spatial genomics technologies, used in academic and translational research, and Xenium has been positioned as a way to extend those approaches to large, clinically derived tissue cohorts where preserving tissue context is essential.

PharosAI aims to generate spatial molecular data, linking gene activity to where different cell types sit within the tissue. According to the press release, Xenium was chosen for its ability to support reproducible analysis across large numbers of archived tissue samples, and because its custom-designed gene panels can be aligned with the molecular features most relevant to specific cancers.

The project is not intended to stop at data generation. PharosAI plans to develop AI models and analysis tools in parallel, so that the datasets produced can be interrogated and reused without extensive processing, a step that often slows down large-scale cancer studies.

The resulting combination of datasets, models, and tools is intended to form a shared research resource, with secure access planned for external academic and industry users. The endpoint really is about supporting multiple future studies rather than a single predefined research question.

Comparable efforts: building large, multimodal cancer atlases

PharosAI’s approach, combining large cohorts of archived clinical tissue with multimodal molecular data and spatial profiling, places it alongside a still relatively small number of initiatives aiming to build reusable cancer data infrastructure.

One of the closest comparators is Owkin’s MOSAIC program, which was launched in 2023 as a $50 million effort to assemble a large-scale spatial omics atlas of cancer. MOSAIC is built around thousands of tumor samples sourced from academic and clinical partners, with the goal of linking spatial molecular data to computational analysis for biomarker discovery and therapeutic research. Like PharosAI, the project is positioned as a long-term resource, although it is driven by the French AI company Owkin rather than a public-sector consortium.

A public-sector reference point is the Human Tumor Atlas Network (HTAN), a U.S. National Cancer Institute initiative launched under the Cancer Moonshot. HTAN brings together multiple research centres to generate multidimensional atlases of tumor evolution, integrating molecular, cellular, and morphological data across cancer types. While HTAN is more decentralised in execution as it is a network of multiple research centers, it does share the objective of following shared standards so the data can be combined and reused for large-scale computational analysis.

In the U.K., PathLAKE is also worth mentioning. It is a national program focused on building AI-ready digital pathology infrastructure within the NHS. Although PathLAKE is centred on imaging data rather than spatial transcriptomics, it similarly aims to enable large pathology archives for research and innovation. Notably, Owkin joined the PathLAKE consortium in 2023.

Today, Zurich-based BC Platforms also announced a partnership with OmicsBank to expand access to large-scale clinical, imaging, and multi-omics datasets drawn from hospital networks in India and the Middle East. While focused on aggregating and enabling access to existing healthcare data rather than extracting new molecular profiles from tissue, it’s a similar attempt to address data fragmentation.

With computational and AI methods becoming more mature, progress is increasingly constrained by access to large, consistent, and well-structured datasets. PharosAI’s bet is that combining NHS pathology archives with multimodal profiling, while developing analytical tools in parallel, can help address that bottleneck. Whether the program delivers on that promise will depend on execution: data quality, standardisation, and real-world reuse of what it manages to build beyond the founding institutions.