OmniGate
I built OmniGate, a tool that answers questions about real single-cell perturbation data in plain English, and shows its work.
Core trick
Model-written Python runs in a sandbox on the real cell data, and predictions are scored by hiding known perturbations and predicting them blind.
Description
I built OmniGate, a tool that answers questions about real single-cell perturbation data in plain English, and shows its work.
"Analyze" takes a question, writes the analysis in Python, runs it in a sandbox against the actual cells, and returns the figure, the statistics, and the exact code that produced them.
"Predict" covers experiments nobody has run: it finds the closest perturbations that were measured, blends their real signatures into an estimate, then audits itself by hiding each known perturbation, predicting it blind, and scoring the result against the truth it withheld.
Why it matters is because single-cell analysis is locked behind knowing scanpy, and most tools that unlock it for non-specialists do so by letting a language model narrate plausible-sounding results. OmniGate never lets the model produce a number. The model writes code, the sandbox runs it, and the answer is whatever the data says.
README excerpt
# OmniGate  **Ask real single-cell data a question in plain English. OmniGate writes the analysis, runs it on the actual data, and shows you the figure, the numbers, and the code it ran. Then it predicts perturbations nobody has measured, and grades its own reliability instead of asking you to take its word for it.** OmniGate has two modes: - **Analyze** turns a plain-English question into real Python, executes it against a real single-cell dataset in a sandbox, and returns a computed figure, computed statistics, and a short interpretation. No canned answers: the chart and every number come from code that actually ran. - **Predict** estimates the effect of a perturbation (a gene knockout or a drug) that was never measured, by blending the real expression signatures of the closest perturbations that *were* measured. It then shows a leave-one-out self-audit so you can see how far to trust it. The rule the whole project is built on: **every number shown is computed from real data.** The model chooses what to compute and which analogs to reason from; it never invents a measurement. --- ## Screenshots **Analyze** and **Predict** share one design language: a soft periwinkle-to-navy gradient that reacts to the cursor, an editorial serif, and frosted panels. | Analyze | Predict | |---|---| |  |  | A finished analysis. T