SimGEXPwMotifs

SimGEXPwMotifs models transcription factor (TF) motif activities to infer their linear influence on gene expression using Bayesian Linear Mixed Models that account for sample correlations.

Key Features:

• Motif Activity Modeling: Models the influence of TF motifs within cis-regulatory elements on gene expression by simulating expression as a linear function of motif activities.
• Bayesian Linear Mixed Models (BLMM): Extends Ridge Regression into a Bayesian Linear Mixed Model framework to accommodate correlations between samples (e.g., same cell line or tissue).
• Simulation and Performance Analysis: Uses simulation studies to compare BLMM and Ridge Regression, showing BLMM outperforms Ridge when unexplained noise is uncorrelated across samples but not when noise shares covariance with the motif-explained signal.
• Mathematical Insights: Provides mathematical explanations for the conditions under which BLMM outperforms Ridge Regression and when no performance gain is expected.
• Real Data Application: Applies the approach to two representative real datasets and reports that up to ≈40% of gene expression signal can be explained by motifs in linear models, with similar covariance structures between noise and motif signal removing BLMM advantages.

Scientific Applications:

• Gene regulatory network analysis: Infer TF motif activities and quantify their contributions to gene expression in studies of gene regulatory networks.
• Analysis with sample dependencies: Model datasets with correlated samples (for example, replicates from the same cell line or tissue) by explicitly accounting for sample covariance.
• Cellular identity and behavior studies: Support investigations into regulatory control of cellular identity and behavior by estimating linear motif-driven components of expression.

Methodology:

Simulate gene expression where expression is determined by linear effects of TF motifs; fit Bayesian Linear Mixed Models as a Bayesian extension of Ridge Regression to account for sample correlations; perform simulation studies comparing BLMM to Ridge Regression; provide mathematical analyses; apply methods to two real datasets.