Autoencoder-Transformed Transcriptome Improves Genotype-Phenotype Association Studies.

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Transcriptome-wide association study (TWAS) links genetic variations to phenotypic traits using gene expression data.
• Predicting gene expression is crucial for TWAS, but low heritability in many genes limits model performance.
• Existing TWAS methods struggle with noise and nonlinearities in gene expression data.

Purpose of the Study:

• To propose AE-TWAS, a novel method that enhances TWAS by denoising transcriptome data prior to analysis.
• To improve the heritability and connectivity of gene expression data for more accurate genotype-phenotype association mapping.
• To identify functionally relevant genes for diseases through improved TWAS.

Main Methods:

• Developed AE-TWAS, incorporating an autoencoder (AE) based data transformation step before standard TWAS.
• Split the transcriptome into co-expression modules and used AE to reconstruct data within each module, effectively removing noise.
• Applied the transformed data to downstream TWAS for genotype-phenotype association analysis.

Main Results:

• AE-TWAS transformation significantly increased expression heritability, especially for genes with initially low heritability.
• The denoised transcriptome data exhibited enhanced connectivity within co-expression modules.
• AE-TWAS improved TWAS performance and identified disease-relevant hub genes with greater functional significance.

Conclusions:

• AE-TWAS effectively denoises transcriptome data, enhancing gene expression heritability and module connectivity.
• The proposed method improves the accuracy and power of TWAS for genotype-phenotype association studies.
• AE-TWAS facilitates the discovery of functionally relevant genes and potential disease biomarkers.