py-URD

A Python port of URD (Farrell et al., Science 2018) — branching trajectory inference for single-cell RNA-seq via flood-based pseudotime + biased random walks.

• Pure NumPy / SciPy implementation
• v0.1 ships the core algorithmic pipeline:
  • floodPseudotime — probabilistic BFS over the diffusion-map transition graph
  • floodPseudotimeProcess — aggregate runs into a per-cell pseudotime
  • simulateRandomWalksFromTips — biased random walks for lineage attribution
  • processRandomWalks — collapse walks into per-cell visit frequencies
• Pseudotime parity vs R URD on Guo qPCR data (428 cells): Spearman 0.985, Pearson 0.992 (threshold 0.80).

Install

pip install pyurd-bio

(module name is pyurd; the PyPI distribution name pyurd was taken, so this package ships as pyurd-bio.)

Quick-start

import pydestiny, pyurd
# 1. Diffusion map for the transition matrix
dm = pydestiny.DiffusionMap.fit(expression, sigma="local", n_eigs=10)
# 2. URD object wrapping cells + transition matrix
urd = pyurd.URD(cell_names=cell_ids, transitions=dm.transitions,
                dm_eigenvectors=dm.eigenvectors, dm_eigenvalues=dm.eigenvalues)
# 3. Flood pseudotime from a cluster of root cells
floods = pyurd.floodPseudotime(urd, root_cells=root_ids, n=50,
                                minimum_cells_flooded=1, seed=42)
pt    = pyurd.floodPseudotimeProcess(floods, max_frac_NA=0.9)["pseudotime"]
# 4. Biased random walks from tips
walks = pyurd.simulateRandomWalksFromTips(
    urd, pseudotime=pt,
    tip_cells={"lineage_A": [cell_a1, cell_a2], "lineage_B": [cell_b1]},
    n_per_tip=1000, seed=42)
visit_freq = pyurd.processRandomWalks(walks)

Function map

Known limitations (v0.1)

1. Tree-building deferred (buildTree, assignCellsToSegments, divergence statistics) → v0.2.
2. Tip selection algorithms (tipPotential, clusterTipPotential) deferred → v0.2.
3. Force-directed tree layout deferred → v0.2.
4. Random walks in v0.1 use only the forward/back biasing by pseudotime; R's URD also supports a step-size weighting and a more nuanced cell-distance threshold that we will add in v0.2.
5. dropseq DGE preprocessing, NMF doublets, batch correction, gene cascades, impulse models — all deferred to v0.3+ (these are large standalone modules in URD that we won't reproduce in Python).

Citation

Farrell, J. A. et al. Single-cell reconstruction of developmental trajectories during zebrafish embryogenesis. Science 360, eaar3131 (2018).

License

MIT.