The Evidence Layer for Scientific AI

 Quick Explanation Copied

 Long Explanation

Evidence across biological scales

Evolutionary starting point: Sponge transcriptomic data identified up to 18 cell types, including cells combining digestive, neuronal-like, and immune-like expression programs. The authors interpret this as consistent with ancestral multifunctional cells later undergoing division of labor; however, transcriptional resemblance is not functional proof, and species and sampling details are incompletely specified. Sponge cell-type evolution

Regulatory structure: In approximately 90,000 nuclei from three flatworm species, single-nucleus RNA and ATAC sequencing was interpreted as showing that cell-type families retain conserved motif vocabularies while using different motif combinations across species. This supports a model in which cell identity is constrained by conserved regulatory components but diversified by combinatorial genome accessibility. Cross-species sequence divergence and dependence on computational models remain important uncertainties. Regulatory evolution of cell types

Organization from molecules to tissues and circuits

Genome architecture: Dip-C reconstructed three-dimensional genome contacts in 17 GM12878 cells and 18 peripheral blood mononuclear cells. Median contacts per cell were 1.04 million and 0.84 million, respectively; the study reported differences between cell types and between alleles. These are measurements of contact detection and reconstructed organization, not direct demonstrations that every structural difference causes a specific cell function. Single-cell genome architecture

Tissue architecture: High-parametric imaging of more than 2 million cells across five developmental timepoints, spanning 6–13 post-conception weeks, mapped protein-defined cell composition, proliferative states, and spatial adjacency in developing human lung. The principal limitation is that each timepoint came from a single donor and the 30-plex antibody panel may omit cell types or markers. Spatial organization in developing lung

Selective connectivity: In adult mice, monosynaptic tracing showed that pallidal neuronal subtypes do not connect uniformly to striatal cell types. FoxP2-positive arkypallidal neurons contributed 44.7% ± 1.8% of labeled pallidostriatal neurons in the baseline experiment, whereas Nkx2-1-positive/PV-negative and Nkx2-1-positive/PV-positive groups contributed 33.4% ± 2.7% and 12.8% ± 1.8% (n=4). FoxP2-positive neurons were especially represented among inputs to indirect-pathway SPNs and cholinergic interneurons, while prototypic populations were prominent among inputs to PV and SOM interneurons. Tracing establishes anatomical association, not synaptic strength or causal behavioral control. Cell-type-specific neural circuits

What the evidence supports—and does not

The strongest synthesis supported by these records is hierarchical: cell-type identity involves molecular programs and three-dimensional genome organization; cells acquire specialized roles; and those roles become spatially and functionally organized into tissues or selective circuits. This is a cross-study synthesis rather than a demonstrated causal chain. The evidence is heterogeneous—sponge and human studies use different species, tissues, assays, and definitions of “cell type”—so it should not be read as proof that all cell types evolve by the same mechanism. Direct tests combining lineage history, perturbation of regulatory motifs, protein localization, and functional physiology would be needed to distinguish correlation from causation.