backed by Combinator

Spatial proteomics from H&E

Lattice predicts spatially resolved protein-marker maps from routine H&E, delivered as virtual mIF channels. It is the first model in our missing data layer for biology.

Loading interactive slide...
A live Lattice prediction on a public TCGA colorectal H&E. Drag to pan, scroll to zoom, toggle predicted protein channels. No sign-in.

Lattice

Spatial proteomics from a slide you already have.

Each model output is a virtual multiplex immunofluorescence channel predicted at the resolution of the underlying image. Pan-cancer and pan-tissue.

No extra tissue

The prediction is computational. Precious archival blocks stay intact and go straight back on the shelf.

Whole cohorts

Runs across an entire archive, not a handful of chosen sections, so retrospective studies become viable again.

Pan-cancer

Trained across a broad span of tumour and tissue types rather than a single indication and a fixed marker panel.

Evidence

Validated at every scale.

Lattice was developed on thousands of tissue regions and millions of image patches, drawn overwhelmingly from cancer tissue and spanning more than a hundred protein markers. Pan-cancer and pan-tissue by construction, rather than tuned to one indication and a fixed panel.

  1. Pixel

    Predicted intensity against the real stain, pixel for pixel, on tissue the model never saw during training.

  2. Cell

    Cells segmented and typed from the prediction, then matched against the same cells typed from real multiplex.

  3. Region

    Marker composition and tissue organisation across a whole slide or tissue core.

  4. Population

    Predicted phenotypes carried across entire cohorts into stratification and survival analysis, where the readout is a population-level outcome.

Send slides. Get answers back.

We run this end to end as a dry lab. There is nothing for your team to install, validate, or staff.

  1. 01

    You send H&E

    Digitised slides from a trial cohort or an archive, straight from the scanner you already run.

  2. 02

    We run Lattice

    Predicted spatial proteomics across your cohort, then cell segmentation, phenotyping and tumour microenvironment characterisation on top of it.

  3. 03

    You get a readout

    Structured spatial data and a written analysis aimed at the biomarker question you actually asked.

For cohorts that cannot leave your institution, we deploy and run Lattice entirely within your environment.

What partners use it for

Retrospective cohorts

Archives that were never intended for spatial work, profiled without consuming a single section.

Patient stratification

Group a cohort by spatial phenotype and see how the tumour microenvironment differs between them.

Target and mechanism

Look at expression and tissue organisation across a population where the assay was never run.

Triage before the wet lab

Decide which samples justify the cost of real multiplex work, instead of guessing at the shortlist.

Beyond Lattice

The missing data layer for biology.

A patient should be dozens of modalities deep. Almost none are. Cohorts get built around the handful of assays that were affordable at the time, and the measurement that would have answered the question is the one nobody took.

Every assay is a projection of the same underlying biology. Genomics, transcriptomics, proteomics and the slide itself are different instruments reading one system, so a projection you never captured can often be inferred from the ones you did. Strand AI reconstructs what was never assayed, so that a patient record can be read whole.

Lattice predicts spatial proteomics from H&E. Other modalities follow. Each is measured on whether it improves a downstream result like biomarker prediction or patient stratification.

Research use only (RUO). Not for use in clinical or diagnostic procedures.