Network & permutation
Genes in network
Find optimal orderings
Find all orderings that minimize μ_E for this network.
Searching…

Seleccionar método de búsqueda

🔎 Exhaustiva (garantizada)
Evalúa ordenamientos. Encuentra todos los óptimos globales con certeza.
🌲 Branch & Bound (garantizado)
Backtracking con poda por cota μ_E parcial. Garantiza todos los óptimos globales con menos evaluaciones que exhaustiva cuando la poda es efectiva.
🧬 Algoritmo genético (heurístico)
pop=300, gen=800, OX1+swap. Devuelve un pseudo-óptimo; para redes pequeñas puede coincidir con el óptimo global.
μ_E counts configurations that expand under f in the p-adic metric — biologically: destabilizing directions.
μ_A counts configurations that contract (attract) — stabilizing basins.
μ_I counts indifferent configurations.
Invariant: μ_E + μ_A + μ_I = (N−1)·pN
Results
Select a network and press Compute to see the stability profile.

§1 p-adic balls

A p-adic metric on a set of Boolean (or p-state) gene configurations measures distance by looking at how many leading digits of two configurations agree when written in base p. Two configurations that share the same first n digits are within distance p−n of each other — they live in the same ball of level n.

Geometrically, p-adic balls are nested: every ball of level n is partitioned into exactly p sub-balls of level n+1. This gives a rooted tree structure (the Cayley tree) that mirrors hierarchical gene regulatory logic. A permutation A of the gene indices determines which gene coordinates appear first, i.e., which biological variables define the hierarchy.

dA(x,y) = p−j where j = min{i : (Ax)i ≠ (Ay)i}

§2 The three measures μ_E, μ_A, μ_I

For each ball B at level n, we ask: do the images f(B) under the Boolean transition function spread further apart (expansion), stay at the same scale (indifferent), or cluster into a smaller ball (contraction/attraction)?

Formally, the diameter M of the image f(B) in the A-basis is the first position where image vectors disagree. Then:

M < n → Expansion (E)  |  M = n → Indifferent (I)  |  M > n → Attraction (A)

The weighted counts μ_E, μ_A, μ_I assign each ball a weight equal to its number of configurations (pN−n), yielding a probability-like measure over the full state space. The fundamental invariant is:

μ_E + μ_A + μ_I = (N−1) · pN

Biologically: a network with high μ_A relative to μ_E is globally contracting — robust and canalizing. A network with high μ_E is sensitive and potentially multistable.

§3 The optimal ordering π*

The scores depend on the choice of permutation A. The optimal ordering π* is the permutation that maximizes μ_A (or equivalently minimizes μ_E), uncovering the most stable hierarchical structure latent in the network. Finding π* requires searching over all N! permutations; for the networks studied here, it was computed by exhaustive or heuristic search.

The optimal ordering reveals which genes act as the "dominant" hierarchical variables: those listed first in π* control the coarsest partitioning of configuration space into stable attracting basins. This has direct biological meaning: top-ranked genes in π* tend to be master regulators.

§4 The three model organisms

Arabidopsis thaliana — flowering transition (N=13, p=2, 8192 states) The 13-gene network controls the switch from vegetative to reproductive growth. Key genes include FT (florigen), AP1, AG, and TFL1. The optimal ordering π* = [12,3,4,8,5,1,7,2,10,11,9,6,0] places UFO and EMF1 at the top of the p-adic hierarchy, consistent with their role as early-acting developmental switches. Expected scores under π*: μ_E = 26,776; μ_A = 53,880; μ_I = 17,648.
Caenorhabditis elegans — cell fate (N=8, p=2, 256 states) An 8-gene network governing cell fate decisions in early C. elegans embryogenesis (Huang et al. 2013). Genes: cdk-2, cdc-25.1, cul-1, lin-35, cdk-1, fzr-1, cdc-14, cki-1. The optimal ordering π* = [1,0,2,3,6,5,4,7] minimizes μ_E = 956; μ_A = 392; μ_I = 444 (exhaustive search over all 40,320 permutations). The relatively high μ_E reflects the network's role in generating diverse cell fates through divergent dynamics.
Drosophila melanogaster — gap gene network (N=6, p=2, 64 states) The six gap genes (bicoid, hunchback, Krüppel, knirps, giant, tailless) pattern the anterior-posterior axis of the Drosophila embryo. The optimal ordering π* = [0,5,3,4,2,1] (exhaustive search over all 720 permutations) gives μ_E = 64; μ_A = 156; μ_I = 100. High μ_A reflects robust positional information encoded along the embryonic axis.

Upload a custom GRN

You can analyze any synchronous Boolean (or p-state) gene regulatory network defined by a complete transition table. Prepare a CSV file with exactly pN rows and two columns:

# Example: N=2, p=2 (four configurations 0..3) config_from,config_to 0,3 1,2 2,0 3,1

config_from: integer 0 … pN−1, all configurations in base-p enumeration (bit 0 = gene 0, etc.). config_to: the image under f. All pN rows are required; the file must be complete.

Once loaded, go to the Analysis tab, choose "Custom CSV upload…", enter N and p, and use the permutation input to explore different orderings.

Load file
Click to select CSV file
After loading, switch to the Analysis tab and select Custom CSV upload… from the network dropdown. The file is stored in memory for the current session only.