The heads do not vary independently. Measured across development, they co-vary on a low-dimensional coupled manifold — roughly seven effective dimensions — and that number is the key to the whole individual-variation program.<\/p>\r\n\r\n\r\n\r\n
Three measurements, one manifold.<\/strong> The head-coupling structure can be read three independent ways: from the mouse atlas (how the heads co-activate across development), from the sequence front-end (the covariance of predicted regulation), and from the super-enhancer census. All three return the same low-rank object. In the human, adult GWAS genetic correlations between traits approximate it a fourth way — the manifold is visible in population genetics.<\/p>\r\n\r\n\r\n\r\n Why it matters.<\/strong> Measurable adult phenotypes number in the thousands; but they collapse onto the manifold’s few shared factors plus small per-organ residuals. An individual’s genome therefore enters the model as a low-rank adapter spanning roughly seven factors — not thousands of independent knobs. This is the tractability argument for computing individual anatomy at all.<\/p>\r\n\r\n\r\n\r\n The developmental loop.<\/strong> Because the embryo is the cause of the adult, data-rich adult measurements are an indirect assay of data-poor embryonic regulation. Read backward through the model, a measured adult effect identifies which embryonic parameter its locus sets, and in which direction — demonstrated end-to-end with real per-allele effects on the heart, the eye and the kidney, with the liver showing the honest converse: its adult volume signal belongs to metabolism, not to the embryo. The partition separates the ontogenetic timing of gene action instead of assuming it.<\/p>\r\n\r\n\r\n\r\n