Molecular Space-Time: Emergent Prediction from Local Interactions in an Evolved Graph Cellular Automaton
Anthony Nash
Abstract
We investigated how local interactions generate predictive information and complex dynamics by representing molecules as graph cellular automata. In this representation, numerical atom states evolve through interactions along chemical bonds, producing molecular space-time trajectories that provide fingerprints for predicting enzyme inhibition. Ridge regression converts these fingerprints into predictions. In the OpenADMET CYP Inhibition Challenge, blind MA-ST-RAE scores were 1.0092 for fixed-rule backpropagation, 1.0020 for genetic configuration search with backpropagation and 0.9946 for genetic programming with backpropagation. Separate dynamical analyses identified convergence, possible periodic behaviour and persistent complex variation. Two systems exhibited bounded hyperchaotic dynamics, with small initial differences growing in multiple independent directions. Discontinuities in the phase-based cellular automaton transition rule and unresolved fractal dimension leave their classification as strange attractors provisional. These findings support molecular space-time as a framework for studying how local interactions generate predictive information and dynamical complexity.
Read the complete manuscript (PDF, 4.68 MB)
Supporting materials
- Method report, videos and trajectory gallery
- Code and supporting data
- High-resolution manuscript (44.9 MB)
Suggested citation
Nash, A. (2026). Molecular Space-Time: Emergent Prediction from Local Interactions in an Evolved Graph Cellular Automaton. Preprint. https://acnash.github.io/Strange-Matter-Engine/