Evolving GRN–Metabolic Networks: Signal Transduction under Selection

Notebook:
Code and figures are available in the companion Jupyter notebook:
OHARA_Universal_Biology_GRN_Metabolic.ipynb

Introduction

This post couples a gene regulatory network (GRN) with a metabolic network and applies evolutionary search to investigate how signal-transduction–like behavior can emerge under selection for homeostasis. In algorithmic terms, evolutionary algorithms (EAs) form a broad family of population-based stochastic optimizers inspired by natural selection (selection, variation, and inheritance). Genetic algorithms (GAs) are a canonical subset of EAs characterized by fixed-length genotypes, fitness-proportionate or rank-based selection, and variation via mutation and crossover acting on the genotype. Put succinctly:

EA ⊃ GA. Every GA is an EA, but not every EA is a GA.
GA emphasizes recombination and mutation on encoded chromosomes; other EA subfamilies (evolution strategies, genetic programming, differential evolution) use different representations and operators.
The implementation used here is an EA variant that adopts GA-style selection and mutation to evolve GRN and metabolic edges and their couplings; crossover is not essential in this minimal setup. The fitness signal is defined by proximity to homeostatic targets for both gene expression and substrate concentrations.

GRN Diagram:

Model

Composite network. The GRN comprises genes connected by directed regulatory edges: activation (positive) and inhibition (negative), including self-regulation on selected nodes. The metabolic network comprises substrates connected by directed reaction edges, with an explicit exchange pathway modeling bidirectional diffusion or transport with the environment. A signal-coupling from metabolites back to genes closes the loop, enabling substrates to modulate gene expression (a proxy for signal transduction).

Key structural elements:

GRN topology. Directed activation and inhibition, with optional self-loops.
Metabolic topology. Directed reactions and specific exchange links to the environment.
Signal coupling (metabolism → regulation). Selected metabolites influence gene expression levels.

Fitness Objective and Evolutionary Setting

Objective. Maintain both gene expression levels and substrate concentrations near a setpoint of 0.5.
Environment change factor. 0.02 per step to induce mild stochastic drift.
Mutation rate (GRN + metabolism). 0.05 on edges/weights/links.
Signal-coupling mutation probability. 0.10 to accelerate exploration of signaling configurations.
Generational horizon. 1000 generations.

The parameterization balances variation (mutation) and stability (environment), enabling gradual adaptation with occasional regressions and recoveries.

Simulation Procedure

The simulation represents GRN and metabolic networks as matrices and adjacency structures. Each generation performs:

Environmental perturbation. Apply stochastic changes to substrate concentrations; apply exchange on designated metabolites.
Metabolic update. Advance substrates along reaction and exchange edges.
Regulatory update. Map metabolite states to gene expression via signal couplings; update gene levels accordingly.
Mutation. Independently mutate GRN edges (sign/strength/presence), metabolic reactions, and signal couplings; use the higher mutation rate on signaling links to broaden search.
Fitness evaluation. Compute deviation of all gene and substrate states from the 0.5 setpoint; lower deviation corresponds to higher fitness.
Selection and record. Retain fitter configurations (per the EA scheme), log fitness history, GRN state matrices (initial/final), and final GRN/metabolic graphs.

Representative Outputs

Fitness over Generations

The fitness trace over 1000 generations typically exhibits oscillatory behavior: gradual improvements punctuated by sudden drops from disruptive mutations or unfavorable perturbations, followed by recovery as selection reinforces compensatory structures. This pattern is consistent with a rugged fitness landscape under mild environmental drift.

Fitness Diagram

GRN State: Initial vs Final (Heatmaps)

Initial GRN matrices are relatively sparse. After evolution, the final matrices show densification and redistribution of activation/inhibition strengths, reflecting emergent regulatory motifs that stabilize system outputs near the homeostatic target.

Heatmaps

Final Network Graphs

Final GRN and metabolic graphs encode node states by color (low to high expression/concentration). The distributions reveal how regulatory structure biases metabolic flows and how metabolite landscapes, in turn, shape regulatory outputs through the signal-coupling, making the regulation–metabolism feedback explicit.

Output Diagram

Output2 Diagram

Parameter Effects (Qualitative)

Environment change factor (0.02). Encourages gradual evolutionary responses instead of volatile shifts.
Mutation rate (0.05). Provides sufficient diversity to avoid stagnation without destabilizing the population.
Signal mutation (0.10). Prioritizes exploration of signal transduction links, increasing variance but enabling discovery of robust homeostatic configurations.

Interpretation

Selection for homeostasis in a coupled GRN–metabolic system tends to indirectly produce signal-transduction–like pathways. Even without direct optimization of signaling per se, co-evolution of regulation and metabolism yields robust motifs that buffer perturbations. The observed fitness trajectories, GRN densification, and end-state metabolite distributions jointly support this interpretation.

Appendix: Additional Runs

Additional end-state graph sets (GRN and metabolic) in the notebook illustrate stochastic variability. Different mutation histories lead to distinct, yet functional, homeostatic solutions with comparable fitness.

Reproducibility Notes

Generations: 1000
Setpoint (genes and substrates): 0.5
Environment change factor: 0.02
Mutation rate (GRN + metabolism): 0.05
Signal-coupling mutation probability: 0.10
Recorded artifacts: fitness histories; initial/final GRN matrices; final GRN and metabolic graphs

Refer to the companion notebook for full code and figure generation:
OHARA_Universal_Biology_GRN_Metabolic.ipynb

Conclusion

A minimal evolutionary algorithm acting on a coupled GRN–metabolic system, with fitness defined by homeostatic proximity to a common setpoint, reliably yields adaptive regulatory structures and stabilized metabolite landscapes. The results demonstrate how complex regulatory behavior can emerge as a byproduct of selection for stability in interconnected biological subsystems, aligning with the broader view that GAs are one practical instantiation within the more general EA framework used here.

Universal Biology: How to _design life?_