G-perturb
G-perturb re-ranks drug targets from a genome-scale CD4+ T-cell Perturb-seq screen by whether each effect is dependable, not just by how large it is.
Core trick
Applies generalizability theory across guide, donor, and activation-state factors to give each target a dependability coefficient for ranking.
Description
G-perturb re-ranks drug targets from a genome-scale CD4+ T-cell Perturb-seq screen by whether each effect is dependable, not just by how large it is. A large transcriptional effect seen in a single guide or a single donor can be measurement noise, and each false lead costs a validation experiment. I treat every perturbation effect as a measurement in a crossed guide-by-donor-by-condition design and use generalizability theory, a framework from psychometrics, to give each target one dependability coefficient, then rank by effect size weighted by dependability. On the released 44.6 GB pseudobulk this reorders the shortlist: 50 of the top 100 targets by raw effect drop out of the top 100 by dependability. Read within activation states, the coefficient reconstructs the T-cell-receptor module (CD3 complex, ZAP70, LAT) as reliable only in activated cells, with no gene labels supplied. A design study shows the screen is limited by the number of guides, not donors. Every methodological decision was recorded and adversarially red-teamed by a competing model before any result was seen; the analysis regenerates end to end, and the manuscript is submitted to bioRxiv.
README excerpt
# G-perturb **Rank CD4+ T-cell Perturb-seq drug targets by whether the readout is *dependable*, not by how large it is.** G-perturb decomposes each perturbation's effect across guides, donors, and activation states with generalizability theory, then re-ranks targets by how reproducibly their signal generalizes. > Built for **Built with Claude: Life Sciences** (Researcher track, 2026), in partnership with Gladstone Institutes. All work in this repository is done from scratch during the hackathon, per the event rules. > 🧬 **Preprint:** posted on bioRxiv on 2026-07-15, DOI [`10.64898/2026.07.13.738312`](https://doi.org/10.64898/2026.07.13.738312) ([read on bioRxiv](https://www.biorxiv.org/content/10.64898/2026.07.13.738312v1)). Manuscript licensed CC BY-NC 4.0; code Apache-2.0. > **Start here:** [▶ 3-minute demo video](https://youtu.be/zBD30nhal64) · [short summary](./SUMMARY.md) · [full paper (PDF)](./manuscript/main.pdf) · [demo slides](./demo/slides.html) + [voiceover](./demo/read-aloud.md). > 📄 Full design rationale: [`docs/design.md`](./docs/design.md) (canonical design). 📝 Research write-up (methods + results, paper voice): [`docs/writeup.md`](./docs/writeup.md) — the bioRxiv-format LaTeX preprint is in [`manuscript/`](./manuscript/) ([`main.pdf`](./manuscript/main.pdf)). The **method** is a distribution-light generalizability decomposition — facet variance shares plus an irreducible replication floor. The **deliverables** are a reliability-weighted target ranking and