scFactor-Ontology

scfo is a Python package for working with latent factors (gene x factor matrices with per-gene loadings) derived from single-cell transcriptomic analyses.

Documentation

  • Home (this page): FAQ's and installation
  • Tutorials: conceptual overview, end-to-end workflows and examples
  • API: function reference with parameters and return values
  • GitHub repository: https://github.com/mrw2204/scFactor-Ontology/tree/main

Installation:

From GitHub

pip install "scfo @ git+https://github.com/mrw2204/scFactor-Ontology.git@main"

Development install

git clone https://github.com/mrw2204/scFactor-Ontology.git
cd scFactor-Ontology
pip install -e .

FAQ

What is an scfo ontology object?

scfo is a tool that organizes latent factor gene loadings and associated prior-knowledge annotations - for example, calculated using a large-scale single cell atlas - into interpretable, factor-centric ontology resources.

The data structure of an scfo ontology object is based on muon MuData, the multi-modal extension of AnnData, treating factors as observations in .X and .obs. This allows for easy storage of factor loadings (obs = factors x var = genes) and prior-knowledge annotation data (obs = factors x var = feature enrichment scores) as separate MuData modalities.

Typical modalities stored in scfo ontology objects may include enrichment scores/p values from transcription factor regulons, ligand-receptor communication signatures, gene set enrichment analyses (e.g. MSigDB Hallmark Gene Ontology), etc.

What can scFO do?

The centralized, organized design of scfo streamlines many common downstream analyses in latent factorization workflows. scfo functionality includes:

  • de novo calculation of enrichment scores based on feature loadings (e.g. gene-wise regulon matrices)
  • plotting suite for visualization of factor loadings/feature loadings/enrichment scores
  • querying an scfo ontology for genes/sets of interest
  • projecting an scfo ontology onto external AnnData objects (e.g. perturbational data).
  • exporting and reloading scfo ontology objects as .h5mu or as standardized Excel workbooks

How do I use scFO?

See "Installation and Usage" below for quick-start instructions, and the "scFO Overview + General Workflow" page (under Tutorials) for a high-level vignette.

This ReadTheDocs site also includes detailed tutorials for using specific scfo functionalities. First, decide between the following:

  • If you have new factors and prior knowledge loadings: see Tutorials #1, #5, and #6 to learn how to construct your factor scfo ontology.
  • If you want to use a pre-existing scfo ontology: see Tutorials #2, #3, #4, and #7 to learn how to use the scfo ontology to visualize factors/loadings (#2), add new modalities (#4), and project factor scores onto external data (#3 and #7).

Do scfo ontology objects already exist that I can download?

Yes! We have generated scfo ontology objects for general use by the neuro-oncology community, representing cell type-specific latent factors from 1) a 3.3 million-cell atlas of human glioblastoma (GBM), and 2) a 400,000-cell atlas of human low grade gliomas (LGG).

  • Google Drive with example data: https://drive.google.com/drive/folders/1V0PoMmk5PseL3cSd1xrHr_btArUtjnhI?usp=sharing
  • To request access to example data on Google Drive, email Matt Warren at mrw2204@cumc.columbia.edu