RNAseqCovarImpute

RNAseqCovarImpute makes linear model analysis for RNA sequencing read counts compatible with multiple imputation of missing covariates by surmounting the problem of high-dimensional outcome data through gene binning or principal component analysis approaches.

Key Features:

• Gene Binning Method: Randomly bins genes into smaller groups to analyze pseudo-independently, creating M imputed datasets separately within each bin using covariates and log counts per million for genes within each bin.
• PCA-Based Method: Uses principal component analysis via the PCAtools package to avoid binning genes while retaining outcome information in the multiple imputation models, with the optimum number of PCs retained based on Horn's parallel analysis or explained variation thresholds.
• Limma-Voom Integration: Implements limma::voom and limma::lmFit functions for gene expression change estimation on each imputed dataset, with variance shrinking via limma::squeezeVar for the binning method and eBayes for the PCA method.
• Statistical Pooling: Pools results with Rubins' rules to produce combined coefficients, standard errors, and P-values, followed by false discovery rate adjustment for multiplicity.

Scientific Applications:

• RNA Sequencing Covariate Imputation: Enables unbiased multiple imputation of missing covariates in RNA sequencing studies where outcome data must be included in imputation prediction models.
• High-Dimensional Omics Analysis: Addresses the computational challenge of implementing multiple imputation in omics studies with high-dimensional outcome data.
• Differential Gene Expression Analysis: Supports linear model-based differential expression analysis with properly handled missing covariate data.

Methodology:

The gene binning method involves: 1) randomly binning genes into smaller groups, 2) creating M imputed datasets separately within each bin with the imputation predictor matrix including all covariates and log counts per million for genes within each bin, 3) estimating gene expression changes using limma::voom followed by limma::lmFit separately on each M imputed dataset within each gene bin, 4) un-binning gene sets and stacking M sets of model results before applying limma::squeezeVar for variance shrinking, 5) pooling results with Rubins' rules, and 6) adjusting P-values for false discovery rate. The PCA method involves: 1) performing PCA on log CPM values for all genes using PCAtools, 2) creating M imputed datasets with covariates and optimum number of PCs, 3) conducting standard limma-voom pipeline with voom, lmFit, and eBayes on each M imputed dataset, 4) pooling results with Rubins' rules, and 5) adjusting P-values for false discovery rate.