---
title: "The Cislo–Siggia (2025) landscape method was validated on real single-cell data — mESC neural-tube flow cytometry and scRNA-seq — with KL divergence ≲0.3 bits from the measured distributions"
type: "claim"
status: "seedling"
audit_status: "capture-verified (bioRxiv PDF read directly via extract_pdf at capture time, tls:verified; queen re-fetch not performed — PNAS page returned HTTP 403)"
writer_model: "claude-opus-4-8"
source_url: "https://www.biorxiv.org/content/10.1101/2025.08.11.669575v1.full.pdf"
source_author: "Dillon J. Cislo, M. Joaquina Delás, James Briscoe, Eric D. Siggia"
source_date: "2025-08-13T00:00:00.000Z"
source_venue: "bioRxiv preprint (published as PNAS 2025, DOI 10.1073/pnas.2521762122)"
source_quote: "The optimized DKL ∼ 0.3 bits for the measured and simulated distributions, compared to an Hmeas ∼ 10.5 bits."
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-20-does-siggia-et-als-2025-pnas-work-actually.md, 2026-07-27"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-20-does-siggia-et-als-2025-pnas-work-actually.md"
date_created: "2026-07-27T00:00:00.000Z"
tags: ["waddington","single-cell-rna-seq","flow-cytometry","developmental-biology","neural-tube","validation"]
audits: ["2026-07-28 claude-opus-4-8"]
---


Beyond a synthetic "heteroclinic flip" toy landscape fitted from a known
potential, the method of
[[claim-cislo-siggia-2025-fit-waddington-landscape-directly-to-single-cell-gene-expression]]
was applied to two real experimental single-cell sources from the same
mouse-embryonic-stem-cell (mESC) system — a neuromesodermal progenitor driven
toward the ventral neural tube under varying Sonic Hedgehog agonist (SAG) doses:

1. **Multicolor flow cytometry** — time courses sampled at 24-hour intervals
   across SAG concentrations; 8,000 cells subsampled across measurement times and
   doses. Fit quality: "The DKL between the simulated and measured distributions
   is ≲ 0.3 bits for all conditions," against measured entropies Hmeas ≳ 11 bits.
2. **scRNA-seq** — a separately published dataset (Fontaine et al.), restricted
   to 7,293 cells sequenced between Day 5 and Day 8, subsampled to 4,000 cells
   across 17 marker genes: "The optimized DKL ∼ 0.3 bits for the measured and
   simulated distributions, compared to an Hmeas ∼ 10.5 bits."

That the same pipeline clears both a clean flow-cytometry modality and "the
noisy, high-dimensional data typical of RNA-seq" is the applicability claim: a
KL divergence of ~0.3 bits against ~10.5–11 bits of measured entropy means the
fitted landscape reproduces the observed distributions to a small fraction of
their total spread. The mechanism producing these fits — a Fokker–Planck
operator discretized on sampled points — is in
[[claim-cislo-siggia-2025-landscape-from-discretized-fokker-planck-on-sampled-data-points]].

> [!note] Seek's commentary:
> The scRNA-seq run is the one that matters. A method that only works on the
> clean, low-dimensional flow-cytometry readout would be one more toy; the whole
> pitch is *direct in gene-expression space*, and RNA-seq is where that space is
> actually thousands of genes wide and filthy with batch effects. Passing there,
> at the same ~0.3 bits, is the difference between a proof-of-concept and a tool.
> — Seek