Pan-Cancer-Cell-Cell-Comm-Net

The Pan-Cancer cell-cell communication network, data, scripts, results.

Important files:

Data/accessories_v2.rda -- an R data file with the following objects:

• binMap -- bins to categorize samples by cell type score abundance
• ccs -- the cell-cell scaffold (ccs)
• cellECDFS -- the cell score abundance distributions

Data/CellSortedExpressonSets/cell_type_expression_tables.rda

• The cell sorted gene expression source data that was used to create the cell-cell scaffold. includes: blueprintComm, blueprintRare, ramiloski, a96, a97, mab, nov, encode

Data/CellSortedExpressonSets/new_Celltype_map_march8.tsv

• The map of cell names in source data to xCell data.

Key tasks

1. Where's the data?

   • complete data is found in gs://pan-cancer-cell-cell-comm-net
   • data used to construct the scaffold: Data/CellSortedExpressionSets
     • E-MTAB-3819 (Blueprint)
     • E-MTAB-3827 (Blueprint)
     • FANTOM5 downloads hg19.gene_phase1and2combined_tpm.osc.txt.gz (https://fantom.gsc.riken.jp/5/datafiles/latest/extra/gene_level_expression/)
     • E-MTAB-3358 (FANTOM5)
     • GSE22886 (Abbas GPL96 & GPL97)
     • GSE24759 (Novershtern)
     • GSE49910 (Mabbott)
     • E-GEOD-26284 (Encode)
   • the cell-cell scaffold (network): Data/cell-cell-scaffold-narrow-march21.rda
   • xCell cell score estimates: Data/xCell_TCGA_RSEM.txt
   • Processed examples of TCGA sample data used for weighting networks (expression and xcell levels): Data/SampleExpr
   • ebpp (EB++) == GDC TCGA Pan-Cancer RNA expression data
     • RNA (Final) - EBPlusPlusAdjustPANCAN_IlluminaHiSeq_RNASeqV2.geneExp.tsv
     • https://gdc.cancer.gov/about-data/publications/pancanatlas

2. How was the data parsed? (it was primarily parsed to enable cluster computing)

   • probability distributions were constructed using Code/prob_model_v2.R -- this entails either taking a vector of cell scores over all TCGA samples, and computing an empirical cumulative distribution function (ECDF) using the R library stats, ecdf(x) where x is the vector of cell scores. -- or for a given cell type, binning xcell scores into quartiles, and for samples within each bin, fitting an ecdf() on expression values for each gene independently.

3. How was the cell-cell scaffold made?

   • using Code/scaffold.R
   • try it out with Example/build_scaffold_example.sh

4. How was the scaffold weighted for each sample?

   • using Code/compute_weights_v2.R
   • which needs, one R data file per sample with both expression and cell estimates, plus distributions & accessories (as above).
   • try a small working exmaple with Example/weight_network_example.sh

5. How do I get all the data?

   • The complete set of data and distributions is stored in a google bucket, you'll need a google cloud account.
   • See Example/download_from_cloud_bucket.sh
   • Also, much of the data and results are stored in Google BigQuery tables.
   • See Analysis/SQL for examples of how to query the weighted tables and various summaries.

6. How do I make plots?

   • See the notebook in Analysis: plots.ipynb.

Table of contents:

1. Analysis - data and notebooks used in the analysis of networks.

2. Code - the code to build the cell-cell scaffold, weight the networks, and perform permuation testing.

3. Data - objects needed for computing the scaffold and edge weights.

4. Distributions - gene epxression and cell abundance probability distribution examples.

5. Example - small examples showing how the network is built and weighted.

6. Results - edge weights, permuted weights, and the Differentially Weighted Edges (DWEs).