randomized-spatial-pca 0.1.1

Randomized Spatial PCA (RASP): a computationally efficient method for dimensionality reduction of high-resolution spatial transcriptomics data

Randomized Spatial PCA (RASP)

Description

Here, we present Randomized Spatial PCA (RASP), a novel spatially aware dimensionality reduction method for spatial transcriptomics (ST) data. RASP is designed to be orders-of-magnitude faster than existing techniques, scale to ST data with hundreds of thousands of locations, support the flexible integration of non-transcriptomic covariates, and enable the reconstruction of de-noised and spatially smoothed expression values for individual genes. To achieve these goals, RASP uses a randomized two-stage principal component analysis (PCA) framework that leverages sparse matrix operations and configurable spatial smoothing.

Features

• High-Speed Performance: RASP is optimized for fast processing of large spatial transcriptomics datasets.
• Flexible Integration: Seamlessly integrates non-transcriptomic covariates into the analysis.
• Spatially Smoothed Values: Produces reconstructed expression values that account for spatial context.
• User-Friendly: Designed to be accessible for researchers in spatial biology.

Requirements

Dependencies are declared in pyproject.toml and installed automatically. The manuscript results were produced with the following pinned versions (also in requirements.txt):

- numpy==1.26.4
- pandas==2.2.2
- scanpy==1.10.1
- squidpy==1.2.2
- matplotlib==3.8.4
- scipy==1.13.1
- scikit-learn==1.5.0
- python-igraph==0.11.5

The mclust clustering option additionally requires rpy2==3.5.16 and an R installation with the mclust package; all other clustering methods (gmm, louvain, leiden, KMeans, walktrap) are pure Python. gmm is an R-free equivalent of mclust (a tied-covariance Gaussian mixture == mclust EEE).

Installation

Install from PyPI (the import name is rasp):

pip install randomized-spatial-pca

Or install the latest development version directly from GitHub:

pip install git+https://github.com/gingerii/RASP.git

To enable the optional R-backed mclust clustering:

pip install "randomized-spatial-pca[mclust]"   # then, in R: install.packages("mclust")

Usage

See the tutorials folder for an end-to-end example. In brief:

from rasp import RASP
import scanpy as sc

# adata: normalized expression in adata.X, coordinates in adata.obsm['spatial']
RASP.reduce(adata, n_pcs=20, n_neighbors=6, beta=2, platform='visium')
sc.pp.neighbors(adata, use_rep=adata.uns['RASP']['embedding_key'])
RASP.clustering(adata, n_clusters=7, method='leiden')

Covariate integration (two-stage RASP)

RASP can integrate non-transcriptomic covariates (e.g. morphology or histology features). When covariates are supplied, the smoothed stage-1 PC scores are concatenated with the (optionally spatially smoothed) covariates and a second randomized PCA produces the final embedding:

# covariates: an (n_obs, d) array, adata.obs column name(s), or an adata.obsm key
RASP.reduce(adata, n_pcs=20, covariates=['morph_area', 'morph_ecc'])
# the integrated embedding is written to adata.obsm['X_pca_cov']; the key to use
# downstream is always adata.uns['RASP']['embedding_key']
sc.pp.neighbors(adata, use_rep=adata.uns['RASP']['embedding_key'])

Use smooth_covariates (bool or per-covariate list) and scale_covariates to control covariate smoothing and z-scoring.

Citation

If you use RASP in your research, please cite our publication in PLOS Computational Biology:

Gingerich et al. Randomized Spatial PCA (RASP): a computationally efficient method for dimensionality reduction of high-resolution spatial transcriptomics data. PLOS Computational Biology (2024). https://doi.org/10.1371/journal.pcbi.1013759

Article: https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1013759