Generating Multimodal Human Data

First, we receive a tumor resection. From it, we extract a series of ‘core’ samples.

First, we profile tumor samples with our AI models. Next, we choose a large number of regions that are clinically meaningful and representative of overall sample biology to 'core'.

Then, the cores are deposited into multiple TMAs, or 'tissue microarrays', each one containing 130 core samples.

This lets us profile many samples in parallel underneath the exact same experimental conditions.

From each core in a TMA, we gather four forms of data:

H&E to capture the clinical pathology view of the tumor.

Then, on the same core, a custom spatial protein panel to identify cell types.

Whole spatial transcriptomes to capture drug targets and the functional state of cells.

Finally, DNA sequencing to identify genetic alterations.

We make sure they are spatially aligned to reflect original tissue architecture.

We’ve generated a lot of this data. And we’re generating more every single week.

This is, by far, one of the largest of its kind by several orders of magnitude.

1.5

of data

250

cells

104,000

cores

3,500

patients