Similarity Metrics at High Dimensionality - testing for rare cell types

Docs

Documentation and reproducibility are available at:

https://ebony-watson.github.io/scProximitE

Install

pip install scproximite

Note: scproximite was developed using Python 3.8, of you have any issues we recommend using conda and creating a new environment before installing:

conda create --name scproximite python=3.8

conda activate scproximite

pip install scproximite

Run tutorials

1. Get tutorial data from zeonodo: https://zenodo.org/record/6443267 (DOI: 10.5281/zenodo.6443266)
2. Add to the data/framework folder
3. Run jupyter notebook in the tutorials folder

You should now be able to run the tutorial notebooks. Note if you don't have R installed you won't be able to run the notebook that uses R metrics: Proximity_Metrics_R.ipynb.

Datasets

Cellsius

A benchmark dataset of ~ 12,000 single-cell transcriptomes from eight human cell lines. The eight human cell lines were individually profiled by bulk RNA-seq, and mixed in four batches containing mixtures of two or three cell lines each for scRNA-seq profiling.

Batch1: IMR90 and HCT116 (50/50)

• IMR90 is a fibroblast cell line, isolated from fetal lung. Female.
• HCT116 is from human colon carcinoma with epithelium-like morphology. Male.

Batch2: A549 and Ramos (50/50)

• A549 is from human lung carcinoma, cell type is epithelial. Male.
• Ramos cells are from Burkitt’s lymphoma. They are lymphoblasts with B-cell characteristics. Ramos cells are very small (7-10um), so we usually find that they have fewer detected features and lower total count than other cell lines. Male.

Batch3: HEK293 and H1437 (50/50)

• HEK293 is a cell line form human embryonic kidney cells. Female.
• H1437 is from lung adenocarcinoma (i.e. origin is epithelial / glandular). Male.

DA234 (Batch 4): Jurkat, K562, Ramos (40% Jurkat, 55% K562 and 5% Ramos)

• Jurkat is a T-cell lymphoblast cell line. Male.
• K562 is a lymphoblast cell line wih granulocyte/erythrocyte/monocyte characteristics (fairly undifferentiated). Female.

Cell-type annotation:

Correlation of the single-cell to bulk expression profiles was used for cell type assignment, & Single cells were assigned to the cell type correlating most with their expression profile. Cells were excluded if their z-score correlation < 0.2, or if they correlated strongly with more than one bulk expression profile (likely doublets).

Subsets

Datasets are pre-annotated with cell_idx, Batch, cell_line, cell_cycle_phase, gene names etc. and a range of QC metrics (would not necessarily trust). Data is downloaded as an R data object, and were subsequently processed in R. Then convereted from seurat to anndata object using SCEasy.

Final datasets are located in RDM under code/DimensionalityReduction_Aim2/data/Cellsius/:

• Cellsius_Complete_Raw(sceasy).h5ad (Full dataset of all 8 cell lines, only pre-cursor filtering)
• Cellsius_Subset1_Raw(sceasy).h5ad (Subset 1 dataset of all 8 cell lines, only pre-cursor filtering)
• Cellsius_Subset2_Raw(sceasy).h5ad (Subset 2 dataset of all 8 cell lines, only pre-cursor filtering)
• subset1_sce_cleaned(SCEeasy).h5ad (Subset 1 dataset of all 8 cell lines, pre-cursor + some additional filtering)
• subset2_sce_cleaned(sceasy).h5ad (Subset 2 dataset of all 8 cell lines, pre-cursor + some additional filtering)

None of the datasets have been normalised/transformed/scaled

Filtering:

Precursor (done by authors prior to uploading data publically):

• ≥ 10.5 genes per cell [log2]
• ≥ 12.0 total UMIs / cell [log2]
• ≥ 10% mitochondrial genes

Additional:

• Outliers
• ≥ 3 counts in at least 1 cell

Sourced from:https://zenodo.org/record/3238275#.YWYVKBx_VhE

Paper: https://genomebiology.biomedcentral.com/articles/10.1186/s13059-019-1739-7