For more than a decade, digital pathology has suffered from a strange paradox: a landscape full of powerful individual tools, but almost no cohesive ecosystem. Each project solves one slice of the computational-pathology workflow, but none of them span the continuum from bioinformatics → raw microscopy → WSI → annotation → model training → reproducible analysis → results distribution.
Worse, the tools that do try to cover multiple pieces tend to do so through tight coupling, inflexible architecture, single-machine assumptions, or restrictive licensing. This creates an ecosystem with a multitude of problems- namely:
- Every research group reinvents the same infrastructure.
- Innovators are limited by the platform they start on.
- Extensibility is an afterthought.
- Horizontal scaling is rare or nonexistent.
- Multi-modal workflows (imaging + bioinformatics + ML) require glue code and one-off scripts.
This post introduces the architectural answer we've been building; Cyto, Eosin, and Lysis, three independent, composable systems that can be used together, depending entirely on what your project needs.
No vendor lock-in, assumptions about your compute, your models, your storage layer or your downstream science. These tools are made with clean boundaries, well-defined APIs and infrastructure that respects your autonomy.
Below is the high-level architecture and intended use cases for each component.
1. Eosin: The WSI Layer
A horizontally-scalable, open, modern foundation for Whole Slide Imaging.
Eosin provides the one thing every pathology project eventually needs: fast, reliable, standardizable access to whole-slide imagery. Everything else—analysis, modeling, provenance, histomics—is downstream of this.
Key goals of Eosin:
- Web-native tile server with Svelte/Rust seams intentionally visible.
- A viewer/annotation tool that's simple, predictable, and forkable.
- Designed for cloud or on-prem: Kubernetes as the control plane, S3 for cold storage.
- Real-time visualization of analysis results superimposed over slides.
- Research-grade image enhancement, measurement tools, and export.
Most importantly: If your project only needs WSI ingestion + annotation + serving and nothing else you can stop at Eosin. That is an intended use case. You can fork it, embed it, remix it, or integrate it with the systems you already rely on.
Some groups will want nothing more than this, and that's fine.
2. Cyto: The Bioinformatics/Data Layer
A modern, language-agnostic substrate for computational biology workflows.
Cyto exists because biomedical research as a whole lacks a proper typed, multimodal data substrate—a coherent place for WSI metadata, annotations, ROIs, provenance graphs, feature tables, and bioinformatics.
Tools like Snakemake and Nextflow are excellent at orchestrating analysis pipelines, but they assume you already have a data substrate beneath them. Cyto fills that missing layer.
- Resource access is unified by the
cyto://protocol. - Resources are pushed & pulled using the
cytoCLI/API. - Resources derived from NCBI, Ensemble, etc. use canonical namespaces, e.g.
cyto://gene/human/BRCA2for the human BRCA2 gene. - Users and organizations get their own namespaces, e.g.
cyto://sablecrestlabs/gene/mus_musculus/MM4933112 - Resources associated with published, peer-reviewed papers are given the
Verifiedbadge. - It works with Eosin and Lysis, but does not require either.
Researchers need a centralized repository where they can upload their data and receive same-day verification. Others want access to that data long before NCBI gets around to integrating it. Everyone wants a platform that unifies all of these data, assigns authority to them, and makes it all amenable to serious compute.
Using Cyto without Eosin or Lysis is entirely valid. Many computational biology groups already have their own WSI stack and analysis workflow. Cyto is where their data collaboration is streamlined.
While strong centralization is a feature, you're welcome to run your own. Think of it like running your own institutional Docker registry.
3. Lysis: The Analysis/ML Layer
Model training, inference, batch processing, and workflow automation.
If Eosin is the WSI substrate and Cyto is the data substrate, Lysis is where the science actually happens. It solves the last big gap in computational pathology: coordinated, reproducible, scalable analysis that bridges imaging and bioinformatics.
The niche is clear:
- GPU-aware scheduling.
- Distributed inference pipelines.
- Integration with Cyto for cross-modal workflows.
- Pluggable models (PyTorch, JAX, ONNX, whatever you bring).
- Pluggable analysis backends (PathML, TRIDENT, etc.)
- Seamless orchestration of multi-step workflows.
But, crucially, Lysis does not force you to use Cyto or Eosin. If your institution already has its own WSI storage, its own genomics pipeline, and just needs automated model execution... Lysis can stand completely alone.
Why This Solves the Ecosystem Problem
The ecosystem problem in computational pathology isn't that we lack tools. It's that the tools don't cooperate, don't scale, or don't compose. Cyto, Eosin, and Lysis intentionally avoid that trap.
1. They are independent.
You can deploy one, two, or all three. They are not a "stack"—they are components.
2. They are horizontally scalable by default.
Not by accident, experiment, or marketing—they operate at scale by design.
3. They are built for institutions with constraints.
Institutional constraints are normal: some labs can't expose WSIs, move genomic data, or run cloud GPUs. The architecture accounts for them.
4. The APIs are open and intended for direct use.
No forcing function requires you to rewrite your infrastructure. No abstract machine that requires months of onboarding.
5. Forkability is a feature.
Every component is designed to be a good fork upstream or downstream.
Extend, replace, or re-architect without breaking the rest. Licensing terms are compatible with closed source and commercial projects.
6. The coupling is explicit and negotiable.
Cyto + Lysis is like Docker Registry + Kubernetes: they're better together, but neither requires the other.
Toward a Healthy Ecosystem
The end goal here isn't to "own" computational pathology. It's to fill the void where infrastructure should be.
- Cyto for data.
- Eosin for imaging.
- Lysis for analysis.
If the field is ever going to stop reinventing itself every three years, it needs infrastructure that behaves like infrastructure—not a monolith, not a black box, and not a walled garden.
That's the future these projects are meant to support.
-Tom