VCC 2026 Chronicle · 2026-09-11
Borrowed biology beat invented biology
We moved a genome-wide CRISPRi screen into a cell type it never saw — 272 of 300 targets got real signatures.
immutable copy · lens://c36cb3972c5d98857910f11b493ed65e410e74fba81e3f6e4db44abf294acfe2
In plain words
The Virtual Cell Challenge asks: switch off one gene, predict what the other 18,000 do — in a cell type your model has never seen. Instead of inventing how genes respond, we borrowed a real catalog: Replogle's genome-wide CRISPRi screen, run in K562 cells, moved into H1 hESC. It gave 272 of the 300 targets genuine signatures, and our leaderboard score went from -0.149 to -0.021 — rank 534.New words
- CRISPRi — Turning a gene's volume down with a guide — a dimmer, not a switch.
- Perturb-seq — Silence one gene per cell across millions of cells, and read what changed — a catalog of cause and effect.
Where this comes from
Mapping information-rich genotype-phenotype landscapes with genome-scale Perturb-seq doi:10.1016/j.cell.2022.05.013
Replogle et al., Cell 2022
k006 run evidence — scores, coverage, provenance k006-replogle-prior-validation
Kytos Observatory
Deep science
k006 pairs two perturbation priors: the 2025 VCC Atlas (in-distribution H1 hESC validation cells) where it covers the target, and the Replogle K562 genome-wide Perturb-seq bulk signature otherwise. Per-target deltas are applied to real control cells via log1p transport with attenuation and noise. 272 of 300 targets used real signatures; the remaining 28 fell back to a hand-tuned context-conditioned prior. Result: overall -0.021 (from -0.149), pds 0.265, rank 534/883, ~$0.75 of Modal compute.