CimpleG DNAm benchmarking datasets for cell-type classification and deconvolution

Two large DNAm benchmarking datasets specifically gathered and curated for cell-type classification and deconvolution problems.

It includes a leukocytes dataset and a somatic cells dataset in the GenomicRatioSet format from the minfi package.

These can be easily loaded into R with the readRDS function:

my_data <- readRDS("CimpleG_benchmarking_datasets_2/leukocytes/tidy_leuk_data.rds")

Each dataset includes therein sample data like GEO accession numbers, sample name or ID in their original dataset, cell-type label, one-hot encoded data for each cell-type, preferred train/test splits, and others.

Alternatively, you can also load the individual .csv files. If you choose this option, I recommend using the function fread from the package data.table. Below I briefly describe these (.csv and .txt) files for the leukocytes dataset, the same logic applies to the somatic cells dataset:

• tidy_leuk_data_beta-values.csv

  • Methylation Beta values matrix
• tidy_leuk_data_m-values.csv

  • Methylation M values matrix
• tidy_leuk_data_probe-annotation.txt

  • Note regarding probe annotation
• tidy_leuk_data_probe-metadata.csv

  • Probe metadata matrix (chr and location)
• tidy_leuk_data_sample-metadata.csv

  • Sample metadata matrix (sample ID, cell type labels, one-hot encoded labels, etc.)