Building an optimal predictive model for imputing tissue-specific gene expression by combining genotype and...

Area of Science:

• Genomics
• Systems Biology
• Bioinformatics

Background:

• Accurate imputation of tissue-specific gene expression aids in understanding complex human traits.
• Current imputation methods utilize either genotype or whole-blood expression data, both obtainable from blood samples.

Purpose of the Study:

• To develop an optimal predictive model for tissue-specific gene expression imputation by integrating genotype and whole-blood expression data.
• To evaluate and compare the performance of standalone and combined imputation models across 47 human tissues.

Main Methods:

• Evaluated standalone genotype (GEN) and whole-blood expression (WBE) imputation models.
• Developed and tested combined models, including merged datasets (MERGED) and inverse variance-weighted (IVW) integration.
• Optimized a MERGED model by upweighting whole-blood transcriptome data using a fixed ratio of regularization penalty factors.

Main Results:

• The WBE model significantly outperformed the GEN model across most tissues.
• A specific MERGED model demonstrated noticeable improvement over standalone models.
• The optimal MERGED model leveraged a strategic weighting of whole-blood expression data.

Conclusions:

• Combining genotype and whole-blood expression data can enhance tissue-specific gene expression imputation.
• The effectiveness of combined imputation strategies is highly dependent on the chosen integration method and data weighting.