N-of-one differential gene expression without control samples using a deep generative model.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Differential analysis of bulk RNA sequencing (RNA-seq) data is often limited by the availability of suitable control samples.
• Existing methods may struggle to accurately identify disease-specific gene expression changes without appropriate comparisons.

Purpose of the Study:

• To develop a control-free method for single-sample differential gene expression analysis using bulk RNA-seq data.
• To create a generative model capable of identifying the closest healthy tissue representation for any given disease sample.

Main Methods:

• An unsupervised generative model was trained exclusively on healthy tissue RNA-seq data.
• The model learns a low-dimensional representation of gene expression profiles.
• It identifies the nearest normal profile for a disease sample to enable control-free analysis.

Main Results:

• The proposed method successfully performs control-free, single-sample differential expression analysis.
• In breast cancer, the approach identified marker genes and outperformed a state-of-the-art method.
• Genes identified by the model were enriched in cancer driver genes, suggesting biological relevance.

Conclusions:

• The generative model effectively replaces the need for control samples in differential gene expression analysis.
• In silico closest normal comparison is a more advantageous approach than using traditional control samples.
• This method enhances the identification of biologically significant genes in disease contexts.