From 10X Genomics to Takara Bio: what LatchBio’s massive spatial atlas means for biotech

In a major leap forward for AI-driven discovery in the life sciences, LatchBio has launched what it calls the largest open-source human spatial transcriptomics atlas to date. Comprising over 25 million curated cells, the spatial dataset covers 45 distinct tissue types, 63 disease categories, and 11 leading spatial omics technologies, creating a foundational resource for researchers building statistical models of human biology.

The release also includes an agentic curation toolkit and a publicly accessible data portal at console.latch.bio/data-portal, enabling scientists to interact with annotated datasets derived from assays such as 10X Genomics Visium, Bruker CosMx, Takara Bio Seeker, and Vizgen MERSCOPE. Each dataset is structured in H5AD format and mapped to standardized ontologies covering tissue, cell type, disease state, assay platform, and patient identity.

With spatial biology fast becoming a core pillar of computational drug development, translational diagnostics, and AI modeling, LatchBio’s curated atlas offers a data infrastructure layer that removes the need for labs to build internal pipelines from scratch.

How does LatchBio’s spatial atlas expand access to large-scale annotated single-cell data?

Spatial transcriptomics is a fast-evolving subfield of genomics that maps gene expression within the physical context of tissue architecture, allowing researchers to see not just what genes are active—but exactly where and in what cell types. This data is especially useful for modeling tissue heterogeneity, tumor microenvironments, developmental processes, and cell-cell interactions.

By collating datasets from publicly available studies across all major platforms—including 10X Visium and Visium HD, STOmics Stereo-seq, AtlasXOmics DBiT-seq, Element Biosciences AVITI24, and Singular Genomics G4X—LatchBio is filling a critical infrastructure gap for AI labs that require spatial data at scale. Historically, researchers needed to source, normalize, annotate, and integrate data from multiple fragmented sources—a process that could take weeks or months per study.

Each dataset hosted by LatchBio includes layered metadata for assay, tissue, disease, and cell type, with annotations harmonized using ontologies such as Uberon and Disease Ontology (DOID). This structure enables easy filtering, downstream training of machine learning models, and reproducible analysis—key priorities for AI-first biotech teams.

The H5AD format, popular in single-cell workflows, ensures compatibility with Python-based tools like Scanpy and Squidpy, further enhancing the dataset’s usability.

What role does agentic curation play in improving the speed and quality of dataset generation?

A major bottleneck in spatial omics has been the human effort required to curate and annotate raw datasets. LatchBio’s “latch-curate” system claims to reduce this curation timeline by 40x, combining automation with human-in-the-loop design. Instead of relying solely on pre-tagged metadata, the system actively parses full papers, figures, and unstructured supplementary materials to extract contextual information.

This hybrid model ensures that important annotations—such as disease stage, treatment status, or assay resolution—are not lost or misinterpreted during the ingestion process. The system also maintains lineage tracking for each dataset, ensuring that annotations are traceable and version-controlled.

LatchBio CTO Kenny Workman emphasized that scientific modeling is increasingly bounded by data availability, not algorithmic limitations. He argued that large, clean, multi-dimensional datasets like spatial transcriptomes are essential for building generative models of disease mechanisms and cellular behavior.

By adopting an agentic architecture, LatchBio positions itself alongside broader AI trends that prioritize contextual understanding, task automation, and dynamic pipeline orchestration—a shift that institutional investors and research labs are actively betting on.

Why are instrument vendors and solution providers partnering with LatchBio for white-labeled portals?

While LatchBio offers its data portal as a public resource, much of its commercial strategy revolves around white-labeled infrastructure partnerships with assay and kit manufacturers. Companies such as Takara Bio and AtlasXOmics are bundling LatchBio’s portals with their spatial kits, allowing end users to visualize, process, and publish data without setting up their own computational environments.

Colin Ng, Vice President of Business Development at AtlasXOmics, noted that processing high-volume spatial epigenomics data used to take months—often requiring bioinformatics contractors or in-house DevOps teams. With LatchBio’s backend in place, customers can now process raw reads into figures suitable for publication in days.

This integrated model significantly shortens the time from experiment to insight and enables faster repurchasing cycles for the solution provider. In turn, it boosts customer satisfaction and adoption rates—factors that are increasingly driving commercial differentiation in the life sciences tools market.

According to LatchBio, solution providers also benefit from a lower customer support burden, since researchers interact with familiar branded environments while the backend architecture and pipelines are fully maintained by Latch.

How is institutional sentiment shifting toward spatial biology and AI-ready infrastructure?

Over the past 24 months, investor interest in spatial omics has surged, with platforms like 10X Genomics (NASDAQ: TXG) and NanoString Technologies (NASDAQ: NSTG) attracting high institutional visibility—despite operational setbacks. The narrative has shifted from niche academic tools to essential infrastructure for biomarker discovery, companion diagnostics, and precision medicine.

While earlier hype focused on hardware innovation, attention is now turning toward scalable software platforms and cloud-native data portals. LatchBio’s offering fits squarely within this trend, addressing the reproducibility, scalability, and usability pain points that often hinder adoption of spatial assays in industrial settings.

Analysts note that AI-native pharma startups—those designing drugs with end-to-end in silico workflows—are hungry for spatially resolved, annotated datasets to complement structural biology and bulk -omics modalities. As multimodal AI models move from theory to clinical trial planning, infrastructure providers like LatchBio are expected to play a crucial role in making spatial data broadly accessible.

Institutional sentiment around curated infrastructure is also driven by broader themes: cost efficiency in R&D, reduced time-to-insight, and alignment with FAIR (Findable, Accessible, Interoperable, and Reusable) data principles—all of which are becoming prerequisites in modern pharmaceutical innovation.

What does this mean for researchers, AI labs, and future development in spatial biology?

For academic labs, biopharma researchers, and computational biologists alike, the implications of LatchBio’s dataset release are significant. Rather than reinventing curation pipelines or sourcing data piecemeal, teams now have access to harmonized multi-study datasets across various spatial technologies and disease states.

For AI model builders, the availability of standardized inputs dramatically improves the ability to train and benchmark models. Whether the goal is to simulate cellular interactions in cancer, predict drug response in inflammatory diseases, or design tissue-specific delivery vehicles, spatially grounded transcriptomics data offers the level of granularity required for meaningful insights.

Moreover, LatchBio’s white-labeled model provides a clear template for future industry collaboration. Just as cloud-based LIMS (Laboratory Information Management Systems) became standard in genomics, platform-layer solutions for spatial data are now emerging as the next infrastructure layer.

Looking ahead, LatchBio has indicated plans to expand its atlas to include animal models (e.g., mouse and zebrafish), organoid datasets, and possibly plant biology, extending its relevance to synthetic biology, agriculture, and systems medicine.

Its federated data model also aligns with regulatory trends pushing toward data standardization in diagnostics and clinical trials, potentially positioning the company as a backend infrastructure partner not just for research—but also for regulated therapeutics and diagnostics development.

Can LatchBio become the spatial data engine behind next-gen biotech workflows?

As spatial biology moves into its next phase of adoption—marked by clinical application, pharma partnerships, and AI integration—curated infrastructure will be a key enabler. LatchBio is aiming to become the “Stripe for spatial omics,” offering a flexible, branded platform that accelerates insight generation while staying invisible to the end-user.

With venture capital in life sciences increasingly directed toward platform-native solutions rather than standalone tools, the infrastructure model pursued by LatchBio aligns with investor demand for scalability, defensibility, and real-world adoption.

Whether the San Francisco-based company becomes the default standard for spatial data delivery remains to be seen. But by combining technical rigor, AI-native architecture, and real-world partnerships with kit makers, LatchBio has positioned itself at the heart of spatial biology’s data transformation.

For AI labs, biotech startups, and life sciences researchers, the 25 million cell atlas is not just a dataset—it’s a signal that the spatial era of biology is ready to scale.