run_edec_stage_1: This chunk of code from EDec paper Run EDec stage 1...

In bozdaglab/CTDPathSim2.0: Computing multi-omics driven similarity scores between patient tumor samples and cell lines for personalized medicine

This chunk of code from EDec paper Run EDec stage 1 algorithm.

Description

run_edec_stage_1 takes as input the methylation profiles of complex tissue samples, the set of loci with high variability in methylation across cell types, and the number of constituent cell types. It then estimates the average methylation profiles of constituent cell types, and the proportions of constituent cell types in each input sample.

Usage

run_edec_stage_1(
  meth_bulk_samples,
  informative_loci,
  num_cell_types,
  max_its = 2000,
  rss_diff_stop = 1e-10
)

Arguments

meth_bulk_samples	Matrix with methylation profiles of bulk tissue samples. Rows correspond to loci/probes and columns correspond to different samples.
informative_loci	A vector containing names (strings) of rows corresponding to loci/probes that are informative for distinguishing cell types.
num_cell_types	Number of cell types to use in deconvolution.
max_its	Maximum number of iterations after which the algorithm will stop.
rss_diff_stop	Maximum difference between the residual sum of squares of the model in two consecutive iterations for the algorithm to converge.

Details

The first stage of EDec performs constrained matrix factorization to find cell type specific methylation profiles and constituent cell type proportions that minimize the Euclidian distance between their linear combination and the original matrix of tissue methylation profiles. The minimization algorithm involves an iterative procedure that, in each round, alternates between estimating constituent cell type proportions (using estimate_props_qp function) and methylation profiles (using estimate_meth_qp function) by solving constrained least squares problems through quadratic programming. The minimization problem is made tractable by the constraints that methylation measurements (beta values) and cell type proportions are numbers in the 0,1 interval, and that cell type proportions within a sample add up to one. These constraints restrict the space of possible solutions, thus making it possible for the local iterative search to reproducibly find a global minimum and an accurate solution. One key requirement for EDec is that cell type proportions vary across samples. A second requirement is that there must be significant differences across constituent cell type methylation profiles. The latter requirement can be met by providing EDec with loci expected to vary in methylation levels across constituent cell types.

Value

A list with the following components:

methylation

A matrix with average methylation profiles of constituent cell types. Rows represent different loci/probes and columns represent different cell types.

proportions

A matrix with proportions of constituent cell types in each input sample. Rows represent different samples. Columns represent different cell types.

iterations

Number of iterations the method went through before reaching convergence or maximum number of iterations.

explained.variance

Proportion of variance in input methylation profiles over informative loci explained by the final model.

res.sum.squares

Residual sum of squares for the final model over the set of informative loci.

aic

Akaike Information Criterion for the final model over the set of informative loci.

rss.per.iteration

Vector of residual sum of squares for the models generated in each iteration of the algorithm.

bozdaglab/CTDPathSim2.0 documentation built on April 14, 2022, 12:39 a.m.