NextFin news, Researchers based in Munich, Germany, from Helmholtz Munich and Technical University of Munich (TUM), announced on November 3, 2025, the creation of Nicheformer — the first large-scale foundation model designed to reconstruct how cells are spatially organized within tissues. Published in the peer-reviewed journal Nature Methods on October 30, 2025, the model was trained on over 110 million cells, encompassing both dissociated single-cell RNA sequencing and spatial transcriptomics datasets, collectively termed SpatialCorpus-110M. This integration tackles a long-standing challenge in biology: reconciling cellular molecular profiles with their native tissue context.
Single-cell RNA sequencing (scRNA-seq) techniques have revolutionized biological insight by profiling gene expression in individual cells. Yet, these cells are typically dissociated from their original tissue, causing loss of information about their spatial position and neighboring cells. Spatial transcriptomics preserves spatial information but remains technically limited in scalability and resolution. The absence of tools that merge these data types has hindered comprehensive understanding of how cellular identity relates to tissue organization.
Nicheformer addresses this by learning joint representations from both dissociated and spatial data, enabling it to "transfer" spatial coordinates and neighborhood context back onto isolated single-cell profiles. The model consistently outperformed current state-of-the-art approaches, demonstrating that spatial gene expression patterns imprint measurable signals even after cell dissociation. Furthermore, Nicheformer’s architecture allows interpretability, revealing biologically meaningful spatial and neighborhood patterns in its internal model layers, shedding light on AI's ability to learn from complex biological data.
According to co-first author Alejandro Tejada-Lapuerta, a PhD student at Helmholtz Munich and TUM, "With Nicheformer, we can transfer spatial information onto dissociated single-cell data at scale, opening new pathways for studying tissue organization and cellular neighborhoods without conducting additional costly spatial experiments." This advancement reframes the future of computational biology by connecting isolated molecular data with physical tissue architecture.
This development aligns with the emerging computational concept of a "Virtual Cell," a model that simulates cellular behavior and interactions within their physiological environment. Previous models largely treated cells as independent data points, failing to incorporate spatial relationships. Nicheformer is the first foundation model to meaningfully encode spatial organization, reconstructing how cells perceive and influence neighboring cells. The research team also proposes spatial benchmarking tasks to stimulate development of AI systems capable of modeling tissue architectures and collective cellular behaviors — key milestones toward biologically realistic AI-driven models.
From a strategic perspective, Nicheformer’s creation addresses fundamental gaps in tissue biology with transformative potential. The ability to computationally re-map single-cell profiles onto tissue spatial maps accelerates unexplored biological insights at scale, bypassing the technical limitations of spatial transcriptomics. This approach establishes a scalable path for integrative multi-omics data analysis, crucial for decoding cellular microenvironments in health and disease contexts.
The team’s next objective is to develop a "tissue foundation model" learning not only molecular and spatial data but also physical interactions among cells. Such a model will be instrumental for detailed analysis of complex tissue microenvironments, including tumor niches and inflammatory sites, with significant potential to enhance understanding and treatment strategies for cancer, diabetes, and chronic inflammatory diseases. Prof. Fabian Theis, Director of the Computational Health Center at Helmholtz Munich and Professor at TUM, highlighted the impact by stating, "These general-purpose AI models represent cells in their natural context and will revolutionize biomedical research and therapeutic guidance."
In terms of broader impacts, Nicheformer signals a paradigm shift in computational biology and AI integration in biomedical research, enhancing precision medicine and drug discovery pipelines. The quantitative reconstruction of tissue spatial organization from large-scale molecular data will facilitate new biomarker discovery, better disease modeling, and ultimately, improvement in patient-specific therapies.
As the US under President Donald Trump continues to emphasize leadership in AI and biotechnology sectors, innovations like Nicheformer exemplify the critical intersection of AI and health sciences driving the next generation of biomedical breakthroughs. The convergence of foundation AI models with high-dimensional biological data heralds a new era in life sciences, with substantial implications for clinical, pharmaceutical, and academic research worldwide.
Explore more exclusive insights at nextfin.ai.
