freddsle/unpast repository overview

UnPaSt is a novel method for identification of differentially expressed biclusters in gene expression matrix. It searches for gene sets up- or down-regulated in subsets of samples,

UnPaSt repository: https://github.com/ozolotareva/DESMOND2⁠

⁠Usage

# for help:
docker run freddsle/unpast:latest
# example run:
docker run -v /your/data/path/:/user_data/ freddsle/unpast:latest --exprs /user_data/exprs.tsv --out_dir /user_data/out_dir/

⁠Outputs

• <basename>.[parameters].biclusters.tsv - a .tsv table with found biclsuters, where
  • the first line starts from '#' and stores parameters
  • each following line represents a bicluster
  • SNR column contains SNR of a bicluster
  • columns "n_genes" and "n_samples" provide the numbers of genes and samples, respectively
  • "gene","sample" contain gene and sample names respectively
  • "gene_indexes" and "sample_indexes" - 0-based gene and sample indexes in the input matrix.
• binarized expressions, background distributions of SNR for each bicluster size and binarization statistics [if clustering is WGCNA, or '--save_binary' flag is added]

⁠About

UnPaSt is an unconstrained version of DESMOND method (repository⁠, publication⁠)

Major modifications:

• it does not require the network of feature interactions
• UnPaSt clusters individual features instead of pairs of features
• uses 2-means, hierarchicla clustering or GMM for binarization of individual gene expressions
• SNR threshold for featuer selection is authomatically determined; it depends on bicluster size in samples and user-defined p-value cutoff

⁠License

Free for non-for-profit use. For commercial use please contact the developers.