PathMatFac.jl

A matrix factorization model for multi-omic data and biological pathways—implemented in Julia.

Note: this code is under development and subject to change!

Installation

Install in the usual Julia way:

julia> using Pkg; Pkg.add(url="https://github.com/dpmerrell/PathMatFac.jl")

Main Ideas

Suppose you have a big array $A$ of multi-omic data. Each row is a sample, and each column is a particular omic measurement.

This package uses the matrix factorization model of MatFac.jl to model $A = X^T Y$.

• We use biological pathways to regularize the $Y$ matrix. This allows us to interpret each row of $Y$ as a pathway factor.
• We allow biological conditions to regularize the $X$ matrix. I.e., two samples belonging to the same biological condition are expected to have similar attributes.
• The model accounts for distributional differences between omic assays:
  • It includes column-specific shift and scale parameters.
  • It assigns appropriate noise models to assays. E.g., log-transformed mRNAseq data are normally-distributed and somatic mutations are bernoulli-distributed.
• The model accounts for batch effects. If you provide batch identifiers for each sample and each assay, the model will fit batch-specific shift and scale parameters.

Basic Usage

TODO: fill this out as development converges