Gene regulatory networks have hierarchies. Master regulators sit at the top, controlling cascades of downstream genes. Identifying this hierarchy from expression data usually requires time-series experiments, perturbation data, or prior biological knowledge about which genes control which.
Perez-Buendia and Nopal-Coello recover the hierarchy from the raw transition map alone, using p-adic numbers.
Encode each gene's state as a digit in a p-adic integer, where p is the number of states per gene. The regulatory network's transition function — a map from one global state to the next — becomes a function on the p-adic integers. The p-adic norm measures ultrametric distance: two states that differ in the most significant gene (the master regulator) are far apart, while states that differ only in downstream genes are close.
The method minimizes a p-adic stability measure over all possible gene orderings. The ordering that minimizes the measure places master regulators at the most significant positions and downstream targets at the least significant positions. The hierarchy emerges from the optimization — not from biological annotation but from the mathematical structure of the transition map under the p-adic metric.
Applied to Arabidopsis thaliana flower development, the method recovers the known ABC model hierarchy: the correct master regulators (AP1, AP2, AP3, PI, AG) at the top, the correct downstream targets below. No biological input beyond the Boolean transition table.
The p-adic framework works because regulatory hierarchies are ultrametric. Downstream genes can change without affecting upstream genes (short p-adic distance), but upstream changes cascade through the entire network (long p-adic distance). The ultrametric structure of the p-adic numbers matches the ultrametric structure of the regulatory logic. The mathematics fits because the biology is genuinely hierarchical, not because the mathematics was designed for it.