omicsGMF

omicsGMF performs fast matrix factorization for dimensionality reduction, visualization, and imputation of omics data from the general exponential family using the sgdGMF-framework.

Key Features:

• Exponential family support: Handles RNA-seq data (Poisson or Negative Binomial) and proteomics data (Gaussian) without requiring log-scale transformation of counts.
• Deviance optimization: Optimizes deviances from the specified data family rather than using transformed values.
• Covariate correction: Corrects for known sample-level and feature-level covariates, enabling visualization and dimensionality reduction after batch correction.
• Missing value imputation: Imputes missing values after matrix factorization, with particular utility for proteomics data.
• Bioconductor data class compatibility: Accepts SummarizedExperiment, SingleCellExperiment, and QFeature classes as input.

Scientific Applications:

• Dimensionality reduction: Reduces the dimensionality of single-cell and bulk omics datasets while preserving data structure.
• Data visualization: Generates low-dimensional representations of omics data for exploratory analysis.
• Missing data imputation: Recovers missing values in proteomics and other omics datasets through matrix factorization.
• Batch effect correction: Enables downstream analysis and visualization after accounting for known technical covariates.

Methodology:

omicsGMF uses the sgdGMF-framework from the sgdGMF package for matrix factorization.