Skip to content

Oocytezero

OocyteZero asks a concrete question behind a disease that affects up to 1 in 8 women: can today's DNA-reading AI models name the single causal base at an unsolved PMOS (polycystic ovary syndrome) genetic locus?

Code Video Cerebral Valley

Core trick

Scores every measured variant at one locus with frozen DNA sequence models and compares predicted effects against reporter-assay labels.

Tags

healthscience

Description

OocyteZero asks a concrete question behind a disease that affects up to 1 in 8 women: can today's DNA-reading AI models name the single causal base at an unsolved PMOS (polycystic ovary syndrome) genetic locus? It's personal — someone I love lives with PMOS — and it's genuinely unsolved: most of the disease's heritable signal sits in non-coding DNA no one has decoded. Working solo with Claude Code and Claude Science, I built an in-silico saturation-mutagenesis benchmark on 589 measured variants at the DENND1A locus and ran an escalating ladder of 12 model classes — from frozen Enformer/Borzoi to AlphaGenome to the reporter-allele model Malinois. Every one reads regulatory activity, but none resolves the causal allele. Two positive controls prove the pipeline works and the negative is real; a mechanistic diagnosis — tissue gap + data gap + an eQTL-silent locus — explains why. Claude Science independently recomputed every number and caught my own overclaims. It's a rigorous, reproducible NO — and a finding others can build on: a reusable, model- and locus-agnostic benchmark, a bound that stops the field from over-calling causal variants, and the exact wet-lab experiment (a matched-tissue steroidogenic MPRA) that could flip it. Come travel the DNA strand with me → oocytezero.vercel.app
josemacontrerasp-arch/oocytezero★ 10 forksPythonCreated Jul 9, 2026 · last push Jul 12, 2026

README excerpt

# OocyteZero — Activity, not allele

**▶ Live demo — [oocytezero.vercel.app](https://oocytezero.vercel.app)** · click *ACCELERATE* to travel the DNA strand through the whole story.

**📄 Read the evidence —** [17-page report (PDF)](docs/report.pdf) · [full findings log](docs/FINDINGS.md) · [reusable benchmark protocol](docs/BENCHMARK_METHODS.md) · [pre-registered validation plan](docs/VALIDATION_PLAN.md) · [100-word summary](docs/SUMMARY.md)

> **A pre-registered, falsifiable test of whether frozen genomic sequence models can nominate
> causal regulatory variants at PMOS loci — plus the two positive controls that prove the
> resulting negative is real, a mechanistic diagnosis of *why* it fails, and the honest
> region-level deliverable the models actually earned.**

*Built with Claude: Life Sciences — Researcher track, July 2026. Applicant: José (Pepe) Contreras,
TU/e. Functional benchmark: Sankaranarayanan et al., Nat Commun 2025 — STARR-seq in H295R (adrenal)
and COV434 (ovarian granulosa) steroidogenic cells.*

![Activity, not allele — frozen sequence models read which enhancer is active (Borzoi G3 ρ=0.35) but not which base is causal (allele ρ≈0.02), and a purpose-trained supervised head cannot recover it either.](docs/fig_ablation.png)

---

## Why I built this

This one is personal. Someone I love lives with PMOS, and *"we don't really know what causes
it"* turned out to be the honest state of the field — so I went looking.

PMOS (polycystic ovary syndrome, renamed in 20

Team