B-GEX

B-GEX infers multi-tissue gene expression profiles from whole blood gene expression using Bayesian ridge regression and genotype and expression quantitative trait loci (eQTL) information to estimate unmeasured gene expression across multiple tissues.

Key Features:

• Bayesian Ridge Regression Framework: Employs Bayesian ridge regression to model relationships between whole blood gene expression and target tissue expression.
• Feature Selection from Blood Profiles: Extracts low-dimensional feature vectors from whole blood gene expression via feature selection for each target gene and tissue.
• Training with GTEx RNAseq Data: Trained on GTEx RNAseq data from 16 tissues to learn cross-tissue expression correlations.
• Performance Superiority: Outperforms least square regression, LASSO regression, and traditional ridge regression across most tissues by achieving lower mean absolute error, higher Pearson correlation coefficients, and reduced root-mean-squared errors.
• Inference of Tissue-Specific and Non-Tissue-Specific Genes: Infers both tissue-specific and non-tissue-specific genes across tissues using only blood input without requiring genomic features or multiple tissue gene expression profiles.

Scientific Applications:

• Basic Biological Research: Enables inference of gene expression in uncollected tissues to study cross-tissue gene regulation and expression dynamics.
• Cancer Studies: Facilitates investigation of tumor-associated gene expression in inaccessible tissues to inform disease mechanisms and potential therapeutic targets.
• Personalized Medicine: Permits inference of patient-specific tissue gene expression profiles from accessible blood samples to support personalized analyses.

Methodology:

Data preparation: select relevant features and extract low-dimensional whole blood expression vectors; model training: train Bayesian ridge regression on GTEx RNAseq data from 16 tissues incorporating genotype and eQTL information to learn cross-tissue correlations; inference and validation: predict tissue gene expressions from blood and validate against known datasets using mean absolute error, Pearson correlation coefficients, and root-mean-squared errors.