thundeR

THUNDER is a deconvolution method specifically tailored to address unique challenges of estimating cell type proportions in Hi-C data. Our algorithm is a two-step deconvolution approach based off of NMF, which first performs feature selection specific to the type of Hi-C readout, then provides estimated cell type proportions using only informative bin-pairs.

Installing and Running THUNDER

Downloading the GitHub repo

In order to run the code listed in this README, you should clone this repository using the following command from a terminal session.

git clone https://github.com/brycerowland/thundeR.git

R Package Installation

THUNDER can be installed using the devtools R package with the following R code.

if(!require("devtools")){install.packages("devtools")}
if(!require("NMF")){install.packages("NMF")}
devtools::install_github("https://github.com/brycerowland/thundeR.git")
library(thundeR)

Estimating cell type proportions from bulk Hi-C data can be performed using the run_thunder function as follows:

thunder_fit <- run_thunder(path_to_mixture = "path/to/thundeR/tests/testthat/new_cols_test_data2.tsv", 
            n_cell_types = 2, itter = 1,
            subset_mix_out_path = "path/to/thundeR/tests/testthat/outfile.txt.gz", 
            out_init_nmf = "path/to/thundeR/tests/testthat/init_nmf_fit.RDS")

where path_to_mixture is the file path to the input data (see below for formatting requirements), n_cell_types is the number of cell types assumed in the Hi-C mixtures, itter is the total number of itterations to run the NMF algorithm (default is 200), subset_mix_out_path is a required argument to output the subset mixture matrix after an initial feature selection step, and out_init_nmf is an optional argument that is a file path to output the initial NMF object estimated from step 1 of THUNDER.

The run_thunder function returns an NMF object. Cell type proportion estimates are estimated via the get_props function:

proportion_estimates <- get_props(thunder_fit)

Mixture data format

THUNDER assumes that mixtures of Hi-C data are stored in long data format where the first row of the file are column names with the following specifications.

• The first column of the file is feature_name and values in this column describe Hi-C data features.
• The second column is contact_type which describes whether the feature is derived from intra-chromosomal contacts or inter-chromosomal contacts. THUNDER performs feature selection seperately for intra and inter features.
• The remaining columns are assumed to be data from a Hi-C experiment stored in long format, where each column is a seperate sample.