Dependencies

Cellar depends on several Python and R libraries. These can be both installed in a conda environment.

---

View Cellar’s conda environment for a full list of libraries and their versions. Below we describe only the libraries that Cellar depends on explicitly.

Python Dependencies

• dash, dash-bootstrap-components, dash-bio
• numpy, scipy
• scikit-learn, scikit-learn-extra, scikit-image
• pandas, anndata, scanpy
• matplotlib, plotly, python-kaleido, tifffile
• igraph, leidenalg
• umap-learn, pydiffmap
• faiss-cpu, joblib
• zodb
• r-base, r-stringi, rpy2, anndata2ri
• diffxpy, gseapy, pyensembl, bintogene

R Dependencies

• SingleR
• cisTopic
• STvEA

NOTE: We prefer to install R (r-base) via conda as seen above. The few remaining R libraries can be installed by using the install_Rdeps.py script while inside the conda environment.

---

Description

We use Dash open-source for the interface and all the UI components. NumPy is divine, and Scikit-learn is at the core of many of the tools that Cellar uses, including several dimensionality reduction and clustering methods. AnnData is the data structure that we use to store annotations and is also used as a session file. Scanpy is at the core of the preprocessing tools as well as the Ingest integration algorithm. Leidenalg is used for Leiden clustering, which is the default clustering algorithm that Cellar uses, and faiss-cpu is used for fast neighbors computation when constructing the neighbors graph for Leiden. UMAP-learn and pydiffmap are used for dimensionality reduction. Rpy2 serves as an interface for running R functions from within Python. We use diffxpy for differential gene testing and GSEAPY for enrichment analysis. SingleR and STvEA are two other data integration algorithms, and cisTopic is used to obtain cell-by-topic matrices from scATAC-seq data. BinToGene is used to obtain a cell-by-gene matrix from scATAC-seq data.

---