CFCR Cell Fate Control Resources

Models, benchmarks, datasets, module collections, protocols and databases for controlling cell fate — indexed by what each resource lets you do, not by model architecture.

How to read the two badges on every card

Evidence and availability are orthogonal: a well-validated model can be unusable in your setting, and a permissively licensed one can be entirely unchecked.

Evidence — derived, never asserted

Recomputed by QC from the attached evaluation papers and their author overlap with the primary paper. It measures independence only: a model is not promoted for being evaluated repeatedly by the same group. The evaluating papers are listed under each badge, so the derivation is auditable.

  • externally-benchmarked at least one evaluation shares no authors with the primary paper
  • partially-independent an evaluation shares some authors, but under half of its own author list
  • self-reported every evaluation comes from substantially the same group, or there is none

Availability — read it before you read the paper

  • open-source code and weights published
  • open-weights weights published, but check the licence — it may forbid commercial use
  • api-only nothing to download; availability rests on the provider continuing to serve the endpoint

Benchmarks are indexed, never run: CFCR stores no per-model scores.

About CFCR

CFCR is a curated index of the models, benchmarks, datasets, module collections, protocols and databases used to control cell fate. It indexes resources you act with, not resources you observe with: trajectory inference and RNA velocity are out of scope because they describe where cells are going rather than how to send them.

Each resource is filed under the question it answers — its fate_axis — rather than under its model architecture: where on the genome do I act, what do I perturb, how do I represent and score sequence, with what do I build the actuator. The axis is multi-valued, and datasets, benchmarks and protocols inherit the axis of whatever they serve.

The one thing that distinguishes it from a list: evidence_tier is derived, never asserted by a curator. QC recomputes it from the attached evaluation papers and their author overlap with the primary paper, and fails the build if a record claims otherwise. It measures independence only — a model is not promoted for being evaluated repeatedly by the same group — and every evaluating paper is listed under the badge, so the derivation is auditable from this page.

Download

Everything behind this page is open and downloadable. The committed TSV snapshots in db/ are the source the loader and this site both read; the JSON files are what the site itself fetches.

WhatFormatTerms
db/entities.tsv — one row per resource TSVCC-BY-4.0
db/papers.tsv — paper metadata TSVCC-BY-4.0
db/entity_paper.tsv — the evidence joins, with roles TSVCC-BY-4.0
db/links.tsv — outbound resource links TSVCC-BY-4.0
site/data/entities.json — entities with papers and links inlined JSONCC-BY-4.0
site/data/papers.json — papers with their citing entities JSONCC-BY-4.0
site/data/facets.json — the filter vocabulary JSONCC-BY-4.0
the cfcr loader — filtered queries over the snapshots Python packageMIT

The whole repository, data and code together, is one zip archive. The loader installs from the repository: 

pip install git+https://github.com/JimYuhaoWu/cell_fate_control_resources.git
from cfcr import load
load(type="model", evidence_tier="externally-benchmarked")

Terms. Code is MIT; the curated metadata is CC-BY-4.0. Neither covers the resources being catalogued, which keep their own licences, nor the paper abstracts, which remain under their publishers' terms. CFCR stores only open metadata and no full text of paywalled work.

How to cite

No DOI has been minted yet, so cite the repository and the exact commit or release you used — the TSV snapshots are committed, so a commit hash pins the data you saw:

Wu, Y. CFCR: Cell Fate Control Resources. Version 0.1.0. https://github.com/JimYuhaoWu/cell_fate_control_resources (commit <hash>, accessed <date>).

If you used a single record rather than the catalogue, cite the resource's own primary paper, listed on its card. A DOI will be added here when one is minted; until then the commit hash is the stable identifier.

Corrections and additions

Curation is the product, so inclusion is a human decision and every change is reviewable. Two routes, and the first is preferred: