Deep Learning Enables Fast and Accurate Imputation of Gene Expression

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Understanding gene expression is crucial for biological discovery and clinical applications.
• The transcriptome, representing all RNA transcripts, is essential for studying gene function.
• Challenges exist in recovering complete transcriptome data due to missing information.

Purpose of the Study:

• To develop novel deep learning methods for accurate gene expression imputation.
• To assess the performance of these methods against existing imputation techniques.
• To leverage the GTEx v8 dataset for broad applicability across human tissues.

Main Methods:

• Proposed two deep learning models: PMI and GAIN-GTEx.
• Utilized the Genotype-Tissue Expression (GTEx) v8 dataset for training and validation.
• Evaluated imputation performance using predictive accuracy and computational runtime.

Main Results:

• PMI and GAIN-GTEx demonstrated superior performance compared to standard imputation methods.
• PMI excelled in inductive imputation, while GAIN-GTEx was optimal for in-place imputation of protein-coding genes.
• The methods showed robust generalization on RNA-Seq data from three cancer types with varying missingness.

Conclusions:

• The developed deep learning methods offer effective solutions for gene expression imputation.
• These approaches facilitate cost-effective integration of large-scale RNA biorepositories into genomic studies.
• The methods have high applicability across diverse human tissues and disease research.