py-destiny

A Python port of destiny (Angerer et al., Bioinformatics 2016) — diffusion maps + diffusion pseudotime (DPT) for single-cell RNA-seq.

• Pure NumPy / SciPy implementation (no Rcpp dependency)
• 3/3 core algorithmic outputs parity-validated:
  • Eigenvectors Procrustes = 0.916 vs R (threshold 0.80)
  • Eigenvalues Pearson = 0.967 (threshold 0.95)
  • DPT Pearson = 0.974 (threshold 0.85)

Install

pip install pydestiny-bio

(module name is pydestiny; the PyPI distribution name pydestiny was taken by an unrelated project, so this package ships as pydestiny-bio.)

Quick-start

import pydestiny
dm = pydestiny.DiffusionMap.fit(expression, sigma='local', n_eigs=5)
dpt = pydestiny.DPT(dm, root=0)

Function map

Known limitations (v0.1)

1. Plotting deferred (plot.DiffusionMap, plot.DPT, gene-relevance etc.) → v0.2.
2. GeneRelevance not yet ported → v0.2.
3. Censoring / dropout-noise model not yet ported → v0.2.
4. Local sigma uses simple mean of n_local nearest distances; R has more sophisticated weighting that slightly shifts eigenvalues (Pearson 0.97 vs perfect 1.0).

Citation

Angerer, P. et al. destiny: diffusion maps for large-scale single-cell data in R. Bioinformatics 32, 1241–1243 (2016).

License

MIT.