Diatom Biolab: Perturbation Single-Cell Data for Training Virtual Cell Models
A Los Angeles startup that produces model-ready scRNA-seq perturbation datasets, now orderable on the Cypher AI Marketplace.
By Cypher AI ·
WHAT IS DIATOM BIOLAB?
Diatom Biolab is a Los Angeles startup that produces biological training data for AI models, starting with single-cell RNA sequencing (scRNA-seq) of cells under chemical or CRISPR perturbations. Diatom Biolab was founded by Jing Zhang and Qining (Leo) Wang, and they are building the Bio-Server, a platform that runs nanoliter- to picoliter-scale reactions on semiconductor chips. As of October 2026, Diatom's data production service is listed on the Cypher AI Marketplace.
Diatom has been through several early-stage programs, including Blackbird's Foundry (Cohort 3) and the Magnify incubator at UCLA's California NanoSystems Institute. Diatom also took part in The Engine's Blueprint program and won first place in the 2025 UCLA Innovation Challenge startup competition.
WHAT PROBLEM DOES MODEL-READY TRAINING DATA SOLVE?
Model-ready training data addresses a supply problem: virtual cell models need large, consistent perturbation datasets, and most existing lab data was produced to answer one question at a time. A virtual cell model is an AI model that predicts how a cell responds to a perturbation such as a gene knockdown or a drug. Arc Institute's State model, for example, takes a starting transcriptome and a perturbation and predicts the resulting shift in RNA expression.
Perturbation datasets are expensive to make. For its 2025 Virtual Cell Challenge, Arc Institute generated a dedicated dataset of about 300,000 H1 human embryonic stem cells across 300 genetic perturbations. For a team that wants to train or test its own model, producing data on that scale in-house can mean building a full single-cell workflow first, so ordering it can be the faster route. As models get easier to access, we think the data a team can generate itself is likely to be what sets its work apart.
Consistency matters as much as volume. When cells are processed in different runs, technical differences between batches (batch effects) can look like biology to a model. Diatom records detailed process metadata alongside each readout and uses it to calibrate out batch effects, so its datasets stay comparable across runs.
WHAT CAN YOU ORDER FROM DIATOM BIOLAB ON CYPHER AI?
As of October 2026, Diatom lists one service on the Cypher AI Marketplace, Virtual Cell Training Data Production. The service currently covers scRNA-seq datasets across diverse sample types, with chemical or CRISPR perturbations and process provenance metadata. Batches run from 1,000 to 10 million cells. The service is priced by quote, and Diatom sets the design, scale and timeline with you.
FREQUENTLY ASKED QUESTIONS
What is a single-cell perturbation screen? A single-cell perturbation screen applies many different changes to cells, such as CRISPR knockdowns of individual genes or a panel of compounds, and measures how each cell responds. scRNA-seq reads out each cell's RNA expression, so every response is linked to the perturbation that caused it.
How much does perturbation single-cell data cost? Diatom quotes roughly $0.20 per cell for small batches, falling to about $0.01 per cell at the largest scale. Each order is priced by quote once the design and scale are set.
Does Diatom offer data beyond scRNA-seq? Diatom is expanding into epigenomics, cell morphology and multimodal single-cell data. As of October 2026, its Marketplace service covers scRNA-seq.
HOW DO YOU ORDER FROM DIATOM BIOLAB ON CYPHER AI?
You can browse Diatom's listing at cypherbio.ai/services without an account. To request a quote or place an order, create a free Cypher AI account. Diatom sends the quote, and you pay Diatom directly by purchase order, card or invoice.
If you order from inside a Cypher AI project, the request is linked to that project and its ELN entry, so the team can see what each dataset was ordered for.