The MOSAIC initiative aims to connect molecular profiles with tumor microenvironment context and clinical outcomes to enable AI-driven biomarker discovery and robust cancer subtyping that goes beyond any single modality.

Overall, MOSAIC represents a pivotal multimodal infrastructure for discovering and refining cancer biomarkers and subtypes by mapping tumor heterogeneity at scale.

The study operates as a non-interventional multicenter protocol funded by Owkin, with some authors affiliated to Owkin or industry relationships.

An initial public dataset, MOSAIC Window, draws on data from 60 patients across five tumor types, including multi-omics and clinical annotations, and is accessible through the European Genome-phenome Archive.

The project prioritizes rigorous quality control, data management, pseudonymization, and adherence to FAIR data principles to ensure data findability, accessibility, interoperability, and reusability.

The data framework combines spatial transcriptomics, single-nuclei RNA sequencing, bulk RNA sequencing, whole-exome sequencing, histology, and rich clinical metadata to create an integrated, standardized resource.

Four case studies illustrate how integrated data can illuminate heterogeneity, using differential expression, gene set enrichment, and pathway activity scoring across malignant subpopulations.

While the approach is promising, its clinical value hinges on prospective validation showing improved patient outcomes, with ongoing data releases as MOSAIC advances.

Early MOSAIC Window analyses quantify intra- and inter-patient heterogeneity and reveal associations between malignant signaling activity and the colocalization of immune and stromal cells within tumors.

The rationale is that no single technology captures tumor complexity; integrating modalities preserves spatial context, single-cell resolution, and genomic information to reveal clinically relevant patterns.

MOSAIC is a large multi-center cancer study aiming to map intra-tumoral heterogeneity through integrated multi-omics profiling across more than 2,700 samples from eleven cancer types.