{"id":97,"date":"2026-08-23T08:45:28","date_gmt":"2026-08-23T08:45:28","guid":{"rendered":"https:\/\/genetec.io\/?page_id=97"},"modified":"2026-08-23T09:38:24","modified_gmt":"2026-08-23T09:38:24","slug":"the-ncalgm-model","status":"publish","type":"page","link":"https:\/\/genetec.io\/?page_id=97","title":{"rendered":"The NCA+LGM Model"},"content":{"rendered":"\n<p>The architecture has two loops, one inside the other.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" src=\"\/wp-content\/uploads\/control_loop.png\" alt=\"The Cognimed control loop: the asymmetric zygote supplies the feedforward prior W0; LoRA adapters sum to W = W0 + delta-W; the NCA (controller and plant) produces growth waves and tissue morphogenesis as a membrane-potential map; the Genomic LLM senses the Vm deviation and feeds delta-W back to the weights.\" style=\"width:100%;height:auto\" \/><figcaption class=\"wp-element-caption\">The control loop. The zygote supplies the feedforward prior W<sub>0<\/sub>; the inner NCA (controller + plant) grows the tissue; the outer genomic model senses the bioelectric V<sub>m<\/sub> deviation and feeds back low-rank weight corrections &Delta;W. A two-degree-of-freedom feedback controller, implemented in chromatin.<\/figcaption><\/figure>\n\n\n\n<p><strong>The inner loop: a Neural Cellular Automaton.<\/strong> Every cell runs the same small update rule on its own state and its neighbours&#8217; &mdash; the biological substrate of weight sharing. Iterated under deep supervision, the NCA is a tiny recursive reasoning machine: the tissue thinks its way to a shape. The <a href=\"https:\/\/genetec.io\/?page_id=91\">three heads<\/a> are its readouts.<\/p>\n\n\n\n<p><strong>The outer loop: a Large Genomic Model.<\/strong> The genome is literal sequence &mdash; A, T, G, C &mdash; read by a frozen encoder into regulatory weights: which enhancers drive which genes, at what strength, in which chromatin context. The LGM does not generate sequence; it converts sequence into the parameters the inner loop runs on. Individual variation enters here as low-rank adapters on the same weights.<\/p>\n\n\n\n<p><strong>The objective: predict the latent, not the pixels.<\/strong> The model is trained to predict the bioelectric membrane-potential field &mdash; the compressed latent in which anatomy is specified &mdash; rather than to reconstruct every molecular detail. Prediction in representation space rather than pixel space is exactly the JEPA principle, arrived at here from the biology.<\/p>\n\n\n\n<p><strong>Glass box.<\/strong> Unlike a conventional deep network, every layer corresponds to an assay: ChIP-seq measures the attention weights, ATAC-seq the activations, RNA-seq the outputs, patch-clamp the conductances, voltage imaging the latent itself. Run forward from a single cell, the trained model grows a recognizable vertebrate &mdash; the demonstration that the architecture is sufficient. See <a href=\"https:\/\/genetec.io\/?page_id=90\">the papers<\/a>, especially Papers 1, 5 and 6.<\/p>\n\n","protected":false},"excerpt":{"rendered":"<p>The architecture has two loops, one inside the other. The inner loop: a Neural Cellular Automaton. Every cell runs the same small update rule on its own state and its neighbours&#8217; &mdash; the biological substrate of weight sharing. Iterated under deep supervision, the NCA is a tiny recursive reasoning machine: the tissue thinks its way [&hellip;]<\/p>\n","protected":false},"author":0,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"footnotes":""},"class_list":["post-97","page","type-page","status-publish","hentry"],"_links":{"self":[{"href":"https:\/\/genetec.io\/index.php?rest_route=\/wp\/v2\/pages\/97","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/genetec.io\/index.php?rest_route=\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/genetec.io\/index.php?rest_route=\/wp\/v2\/types\/page"}],"replies":[{"embeddable":true,"href":"https:\/\/genetec.io\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=97"}],"version-history":[{"count":3,"href":"https:\/\/genetec.io\/index.php?rest_route=\/wp\/v2\/pages\/97\/revisions"}],"predecessor-version":[{"id":134,"href":"https:\/\/genetec.io\/index.php?rest_route=\/wp\/v2\/pages\/97\/revisions\/134"}],"wp:attachment":[{"href":"https:\/\/genetec.io\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=97"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}