Z-Screen - Investor Briefing - April 2026

Turn new chemistry into searchable biology.

Z-Screen makes combinatorial molecules, profiles their cell-state effects, and keeps every response tied to the recipe that produced it. The pilot is already one of the largest public combinatorial chemistry transcriptomic datasets. More investment can scale the loop to million-compound campaigns, richer cell models, and faster mechanism triage.

615,793RNA profiles in the pilot
142,187Hashed compounds with chemistry embeddings
12 / 4Libraries / cell lines
1,276Calibrated CRISPR-state hypothesis tuples
Why investors should care

The pilot shows a compounding discovery loop.

The current release is evidence that learn-in-a-loop at scale works: make chemistry, read cell state, learn the map, and feed the next campaign.

PLATFORM UNIT

A 50,000-well chip replaces a slower screening workflow.

The chip is the throughput unit. More campaigns can mean more chips, more libraries, and more cell contexts rather than a bespoke automation build for every program.

50KWells per chip
DATA ASSET

Every molecule teaches the model what chemistry did.

Because each RNA profile remains attached to building-block provenance, the dataset becomes a reusable chemistry-to-biology map rather than a one-time hit list.

12Combinatorial libraries
DECISION SPEED

Hits arrive with ranked follow-up hypotheses.

CRISPR-state matching does not prove target engagement. It does turn an unknown phenotypic hit into a prioritized, chemistry-addressable validation queue.

1,714Distinct CRISPR programs linked in the headline scan
OPTION VALUE

The same engine can serve many therapeutic programs.

The platform story is broader than one disease area: combinatorial small molecules today, with a path toward richer image channels, more cell models, and partner-specific discovery maps.

4Cell lines in the public pilot
The five preprints

Five results from one connected dataset.

The cards below are a fast read on what the pilot already observed: paired readouts, learnable chemistry, honest generalization tests, cross-cell transfer, and CRISPR-linked mechanism hypotheses. Click any card for details.

Select a paper card to open the detail page.
PAPER 01 - ActiveSeq
Image and RNA from the same microwell.
ActiveSeq linked microscopy and RNA in the same well. Across a 35-compound control panel, both readouts recovered compound signal, and the combined readout classified held-out wells better than either modality alone.
Open the evidence
PAPER 02 - Generative Chemistry
The screen learns which chemistry moves biology.
Because each profile stays tied to the compound recipe, Z-Screen can learn from chemical building blocks rather than treating hits as isolated events. In the strongest systems, that grammar predicted held-out RNA states.
Open the evidence
PAPER 03 - Generalization Ladder
Generalization becomes measurable.
The ladder separates dense-grid completion from harder chemistry tests. ZEL031 / THP1 showed held-building-block extrapolation beyond nearest-neighbor lookup, while ZEL024 / HEK293 showed high-value completion inside a dense design.
Open the evidence
PAPER 04 - Cross-Cell Transfer
Chemical-response programs transfer across cell types.
Paired cell-line data showed that learned maps recover target-cell RNA response better than direct reuse. The strongest signal appeared at the chemistry-resolved program level, where shared building-block programs transferred strongly.
Open the evidence
PAPER 05 - Mechanism Map
Unknown hits get mechanism hypotheses attached.
Compound RNA signatures were searched against public CRISPR perturbation states, producing calibrated, chemistry-addressable hypotheses. The headline scan linked 1,276 chemical tuples to 1,714 CRISPR programs.
Open the evidence
Additional investments

What scale could unlock next.

The pilot supports the platform thesis, but it was not designed to saturate the opportunity. Focused additional investment should be aimed at the bottlenecks the papers identify: larger chemistry coverage, richer paired readouts, prospective scaffold-hop tests, and matched-cell validation for mechanism hypotheses.

1M+
Compounds with paired mRNA and Imaging
More cells
Disease models, patient cells, and paired landmarks
More images
Target-aware optical channels tied to RNA
Translate
Move predictions to in-vivo validation
Pilot release - April 2026

Read the evidence. Talk to the team.

These draft manuscripts are circulated for feedback ahead of a bioRxiv submission. The full public dataset and analysis scripts are available on Zenodo for independent evaluation.