Cognimed NCA+LGM — the matrices
The trained parameter set of the Cognimed / Cognitive Biology model: an inner Neural Cellular Automaton (the tissue's local update rule) driven by an outer frozen Large Genomic Model (the genome read as regulatory weights), with the bioelectric membrane-potential field as the latent in which anatomy is specified.
The model is deliberately small. Its parameters are read from the genome through public regulatory databases and validated against public spatial transcriptomic atlases — not fitted at scale. This repository contains every matrix needed to reproduce the published results, alongside the open code.
- Code: https://github.com/AlphaFanX/cognitive-biology
- Papers (nine, open access, CC-BY 4.0): DOIs
10.5281/zenodo.20722139,21143761,20746637,20925727,21143016,21221930,21796907,21322049,21791849(all10.5281/zenodo.) - Site: https://genetec.io/ (program overview under Cognimed)
- Author: Miles Bradford Jacobs, genetec.io, Cape Town
Layout
| Directory | Contents |
|---|---|
weights/ |
The outer-model matrices: nca_W0_modules.npz (the feedforward prior W0), gene_partition.json (the GWAS-Catalog-derived kernel/adapter partition of the genome), adapter_table_full.json + adapter_table.json (per-allele trait adapters: each GWAS allele placed at its developmental knob) |
heads/ |
The head-level maps: canonical_map.json (294 body parts matched to canonical reference meshes), tissue_conductances.json (genome-derived ion-channel/gap-junction conductances per tissue), wiring/ (super-enhancer region binders and counts behind the identity heads) |
results/organ_cascade/ |
Every derived result table from the papers: differentiation-clock calibrations, organ placement addresses, the developmental-loop back-propagations (heart/eye/kidney/liver maturation), shape-match scores against dense atlas reconstructions, subtissue regulon namings |
results/grays_scorecard/ |
Per-part anatomical scorecards: all 294 named structures scored in isolation against a Gray's Anatomy checklist |
derived_atlases/ |
Our dense 3D reconstructions built from the public atlases: mouse E12.5/E16.5 (from MOSTA), human (from HESTA), zebrafish (from ZESTA). Derived works — see attribution below |
The honest ledger
The papers keep an explicit account of what is genome-derived and what is anchored: the anterior–posterior organ addresses are derived (single-cell Hox boundary, Spearman 0.81); the dorso-ventral address is partial; the resting-potential magnitude is anchored to one measured value (a gating quantity, not a transcript quantity). See Paper 9 for the full ledger.
Third-party data (not redistributed here)
The raw atlases and databases are third-party resources; fetch them from source:
- MOSTA (mouse Stereo-seq atlas) and ZESTA (zebrafish) — CNGB/STOmics
- HESTA (human embryo Stereo-seq)
- Jadhav 2019 methylation strata — GEO GSE111024
- ABC enhancer–gene predictions — Engreitz lab
- SEdb super-enhancers — sedb release downloads
- GWAS Catalog — EBI full associations download
- AlphaGenome — API access from DeepMind
- BodyParts3D / OpenAnatomy meshes, FaceBase — from their portals (licenses vary; FaceBase mesh must be requested there)
The derived reconstructions in derived_atlases/ are transformative works built
from MOSTA/HESTA/ZESTA; cite the original atlas papers alongside ours when using them.
Citation
Please cite the relevant papers from the series (DOIs above); the foundational
paper is Cognitive Biology: From Systems to Somatic Networks, Tissue Cognition,
Cognitive Morphogenesis, and Tissue Emotions, 10.5281/zenodo.20722139.