Neural Cellular Automata With SIREN Capture Biological Self-Organization Via Morphogen Gradients

Biological development reliably builds complex structure from local cell interactions. How does it do it? Beyond self-organisation itself, a substantial amount of organisational information is offloaded to initial conditions: maternal morphogen gradients and pre-patterns. We model this with a Neural Cellular Automaton paired with a learned coordinate-based generator (SIREN), trained jointly. In our model, offloading information to initial conditions improves robustness, encoding capacity, and symmetry breaking. 🧬 Check it out: