Analysis of Age-related Changes in Intercellular Communication from scRNA-seq

• analysis - main folder with final analysis
• src - folder with utility functions
• test - folder to test and benchmark specific parts of codes
• shinyApp - code for the web app
• old_deprecated - code to keep temporarily (will be removed eventually)

Note: it should not be necessary to run analysis 1-4 each time if the data are already stored.

• analysis_1_Seurat_processing.R - verifies the integrity of Seurat files
  • controls the content of the files
  • checks that QC has been done properly
  • produces some statistical information and plots about each dataset/tissue/cell-type
• analysis_2_LRdb_selection.R - compares the 4 LR databases
  • summary and Venn diagram
  • selection of relevant LR-pairs
  • GO enrichment analysis on the LR pairs
• analysis_3_run_scDiffCom.R - does scDiffCom analysis on all datasets
  • run in parallel over all tissues (one dataset at a time)
  • need to be called each time for each dataset
• analysis_3b_run_diffcom_whole_dataset.R - not functional for now...
  • attempt to consider the entire dataset as a single tissue
  • not working yet, requires too much memory (even on the server)
  • not sure if this is a relevant analysis
• analysis_4_filtering_ora_fpm.Rmd - first downstreamn analysis
  • perform filtering
  • perform overrepresentation analysis
  • perform frequent pattern mining
  • we need to discuss the choice of cutoffs
• TODO
  • add future downstream analysis (general stats, enrichment, more plots, etc)

Note: each script can be run independently, as long as some data files are stored. Additional files can also be added if more tests are needed.

• test_1_scDiffcom.R - checks that scDiffCom returns correct values
  • requires a Seurat file for testing (e.g: "../data_scAgeCom/data_seurat_example.rds")
  • shows typical usage of scDiffCom
  • checks scores and p-values (from independent functions)
  • create plots to compare distributions from permutations
• test_2_compare_to_cpdb.R - check that scDiffCom returns similar values compared to CellPhoneDB
  • requires preprocessed CPDB output
  • shows (deprecated) usage of run_cpdb_from_seurat
  • shows that LR scores are exactly the same between the methods
  • shows that LR specificity p-values are similar (up to randomness)
  • shows high similarity in classifying significant/non-significant CCIs
• test_3_scDiffCom_parameters - compare scDiffCom results for different methods/parameters
  • compares normalization, log-scale and x-sided test
  • checks the classification of each CCI for each choice of parameters
  • checks the shape of the distributions from the permutation tests
• TODO
  • do test with another file, e.g. from Calico
  • add more test related to LR_score cutoff
  • clarify how test_3 can help us to justify our choice of parameters
  • clarify the origin of the strange cases whith very low distribution means