DataRemix

DataRemix performs universal data transformation to optimize inference from RNA-seq gene expression datasets by normalizing data and managing nuisance biological variation and technical covariates.

Key Features:

• R package: Implemented in R for processing and transforming gene expression datasets.
• RNA-seq normalization: Provides an advanced normalization method that manages nuisance biological variation and technical covariates in RNA-seq data.
• Downstream analysis support: Improves inference for gene-correlation network reconstruction and eQTL (expression Quantitative Trait Loci) discovery.
• Generalized SVD reconstruction: Uses a generalized singular value decomposition-based reconstruction that subsumes whitening, rank-k approximation, and removal of top k principal components as special cases.
• Tunable parameters: Transformation is governed by three tunable parameters that adjust the weighting of hidden factors.
• Prioritizes biological signals: Prioritizes biological signal detection without requiring external dataset-specific covariate information.
• Optimization via Thompson sampling: Employs Thompson sampling for efficient parameter optimization, suitable for computationally intensive eQTL analyses.

Scientific Applications:

• Gene-correlation network inference: Enhances detection of co-expression relationships by reducing technical and nuisance variation.
• eQTL discovery: Increases sensitivity for expression Quantitative Trait Loci discovery, including trans-eQTL detection, by adjusting latent factor weighting.
• Human brain transcriptomics (ROSMAP): Applied to the Religious Orders Study and Memory and Aging Project and used to identify a replicable trans-eQTL effect in human brain.

Methodology:

Generalized singular value decomposition (SVD)-based reconstruction that subsumes whitening, rank-k approximation, and removal of top k principal components; transformation parameterized by three tunable parameters to reweight hidden factors; parameter optimization via Thompson sampling.