Gene Regulatory Network Inference

Project Objectives

To develop and validate a computational method for inferring gene regulatory networks from genomic data using a non-local prior Bayesian model as a foundation for high-dimensional variable selection. Other statistical models and variable selection methods (lasso and mombf) are tested and compared.

Set-up

Github

git clone [email protected]:Celine-ZL-Chen/GRN.git

R Packages

• io: read from, write to, plot to in a unified manner
• ggplot2
• glmnet
• mombf
• dplyr
• to be completed

Databases

1. GTEx
2. CGTA
3. DoRothEA
4. to be completed

Data Preprocessing

GTEx ... doro_tf-target_filtered.R

Model Description

Class 1: assume TF expression = TF activity

Model 0 - univariate linear model uni_reg_cdkn1a.R

Model 1 - multivariate linear model multi_reg_cdkn1a.R

Model 2 - multivariate lasso model (glmnet) lasso_cdkn1a.R

Model 3 - multivariate non-local prior model (mombf) mombf_cdkn1a.R

‌Class 2: infer TF activity from target genes' expression

• mean expression of target genes (DoRothEA) as activity of TF [no normalization performed]
• Summerized GTEX data [pick out all activities of the TF]