Compute evolutionary pathways¶

Four families of pathways. Pick the one that matches your question.

shortest_paths (deterministic, single path per target)¶

The shortest path is the path with maximum product of transition probabilities (equivalently, minimum sum of \(-\log P_{ij}\)):

from gpvolve import shortest_paths

ens = shortest_paths(msm, source=0, targets=[31, 63])
for path, prob in zip(ens.paths, ens.probabilities, strict=True):
    print(path, prob)

greedy_walk (deterministic, single max-prob walk)¶

Greedy descent on the transition matrix: at each step, hop to the neighbor with the largest off-diagonal \(P_{ij}\). Stops on a target hit, fixed point, or cycle:

from gpvolve import greedy_walk

ens = greedy_walk(msm, source=0, targets=63)
print(ens.metadata["hit_target"])

dominant_pathways (deterministic, top-k by reactive flux)¶

Decompose the reactive flux from A to B into bottleneck pathways:

from gpvolve import reactive_flux, dominant_pathways

F = reactive_flux(msm.transition_matrix, A=0, B=63)
pathways = dominant_pathways(F, A=0, B=63, top_k=10)

Each returned PathEnsemble has its bottleneck flux in probabilities[0]. The list is sorted by bottleneck flux descending.

sample_paths (stochastic, ensemble with convergence check)¶

See the stochastic sampling guide.