A yardstick for brain organoids, built from real brains

What the work claims

This is a year-one research grant, so the claims divide into demonstrated and proposed. Demonstrated: the Wang lab's earlier tool, Brain and Organoid Manifold Alignment (BOMA), which the grant abstract says integrated eight published single-cell RNA sequencing datasets, and which the 2023 paper shows aligning developmental gene-expression data between brains and organoids at two resolutions, then clustering the aligned samples to flag conserved and organoid-specific programs. The paper reports that human cortical organoids align more closely with certain cortical brain regions than with non-cortical regions, and the tool is available as an open-source web application¹, with a step-by-step cloud protocol published in 2024².

Proposed: the grant's three aims. Aim 1 extends BOMA's transcriptomics-only input to single-nucleus multiomics, under the name BOMAM, to compare brain and organoid development using gene expression and chromatin accessibility measured in the same nucleus. Aim 2 applies gene regulatory network prediction to score the fidelity of current organoid protocols. Aim 3 turns the result into open-source, general-purpose evaluation tools for the brain-organoid field³. The bet is that the gap between an organoid and a real developing brain can be made measurable, and therefore comparable across labs, protocols, and vendors.

How it works

The core idea is manifold alignment, and it is worth explaining concretely because the governance implications ride on it. A gene-expression profile from one sample is a point in a space of tens of thousands of dimensions, one per gene. Samples from related biological states form a lower-dimensional structure inside that space, a manifold. BOMA takes developmental time series from brains and from organoids, performs a coarse global alignment of the two series, then uses manifold learning to refine the mapping locally, so each organoid time point can be placed at a position along the brain's developmental trajectory¹. The output is not a yes-or-no verdict but a correspondence: this 14-week organoid sits near this post-conception week of reference brain.

The proposed extension matters because RNA abundance alone cannot say which regulatory elements drive a gene. Single-nucleus multiomics profiles gene expression and chromatin accessibility from the same cell, linking candidate enhancers to their target genes. The grant leans on an asymmetry in the data landscape: NIH consortia, including the BRAIN Initiative, are generating large single-nucleus multiomic human brain datasets spanning prenatal development to adulthood, while equivalent organoid data are still scarce. BOMAM is designed to plug into that reference as it grows³.

Where a skeptic should push

The single most load-bearing assumption is that alignment to available in-vivo brain data is the right definition of fidelity. That deserves stress from two directions. First, correspondence is not identity. The 2023 paper itself demonstrates the gap: immunofluorescence validation on cortical organoids (n = 3) found the cortical-plate markers SATB2, POU3F2, and PSMB5 significantly enriched in human cortical plate at post-conception week 19 compared with 14-week organoids mapping to the same trajectory segment (p = 0.0038, 0.001, and 0.0011). A molecular clock can place an organoid on the brain's timeline while the tissue itself remains developmentally behind at that placement¹.

Second, the reference is an accident of data availability. BrainSpan-style atlases are dominated by fetal and early-postnatal tissue because that is what exists. An alignment score therefore measures resemblance to a particular, demographically narrow slice of human brain development, and BOMAM will inherit whatever sampling biases the consortia carry. There is also a circularity risk: protocols get tuned until they score well on the manifold, after which the manifold is cited as independent confirmation. Finally, the grant is one year into a five-year project period; BOMAM, the regulatory-network scoring, and the open-source tools are aims, not delivered artifacts. Weight the piece accordingly: a working transcriptomics tool with a real publication record, plus a credible plan, not a demonstrated multiomics standard.

Whose brain defines the fidelity yardstick

For platform access and vendor capability, the significant fact is that verification is getting cheap. BOMA already runs as a free web tool, and Aim 3 commits to general-purpose open-source evaluation. Today a vendor can claim its cortical organoid protocol is "developmentally accurate" and the claim is expensive to challenge; a widely adopted alignment score converts that claim into a number any customer, regulator, or competitor can recompute. That inverts the usual moat: the durable advantage migrates from marketing language to raw data quality, because a mediocre protocol with good metadata can still be scored honestly. It also gives an oversight body something concrete to attach to, a published fidelity score with a defined reference, which is far more auditable than a brochure.

The ethics angle is subtler and, we would argue, more important. The organoid field routinely makes moral-status-adjacent statements of the form "this organoid is equivalent to N weeks of fetal development." Manifold alignment is precisely the machinery that generates such statements, by mapping dish time onto brain time. That means the calibration of the moral-status yardstick is being set by which reference datasets exist, a scientific and funding accident, rather than by any deliberation about which developmental features are morally relevant. A reference atlas heavy on fetal cortex and light on everything else will silently define "advanced" as "fetal-like," which pressures protocols toward fetal mimicry and undervalues deliberately divergent models, such as aged or disease-state tissue, that score poorly precisely because they are not trying to copy a fetus.

The genuine threat is metric fixation: if journals, funders, and procurement offices adopt an alignment score as the validity gate, protocol designers will optimize the score rather than the biology, a textbook Goodhart dynamic, and the reference datasets' demographic gaps harden into everyone else's validity ceiling. The genuine opportunity is a shared, cheap audit layer: the first time a third party can price a fidelity claim without trusting the seller. Whether that layer becomes a public utility or just another leaderboard depends on choices this grant does not make, including who is represented in the reference and who governs score thresholds.

The bottom line

Established: transcriptomic manifold alignment between brains and organoids works, is published in a peer-reviewed methods journal, and is available as an open web tool. Hypothesis: the same approach extends to single-nucleus multiomics and yields a defensible, general-purpose fidelity score for brain organoid protocols. What would confirm it: BOMAM released, adopted outside the originating lab, and shown to predict an independent readout of developmental maturity, ideally functional or electrophysiological, rather than another molecular one. What would break it: alignment scores proving insensitive to known protocol failures, or reference datasets so skewed that scores tell you more about the atlas than the organoid. The yardstick is coming; the open question is whose brains it will be calibrated on, and who decides what counts as close enough.