CONTRASTIVE LATENT VARIABLE MODELING WITH APPLICATION TO CASE-CONTROL SEQUENCING EXPERIMENTS.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• High-throughput RNA-sequencing (RNA-seq) is crucial for understanding cellular states.
• Current differential expression analyses often overlook transcriptional correlation and complex gene expression shifts.
• There is a need for methods that capture low-dimensional gene expression structures and collections of co-changing genes.

Purpose of the Study:

• To propose contrastive latent variable models for count-based RNA-seq data.
• To create a more comprehensive analysis of differential gene expression.
• To develop a framework for testing global and subset-specific expression changes.

Main Methods:

• Developed contrastive latent variable models tailored for count data.
• Integrated baseline variation modeling to disentangle transcriptional sources.
• Created a model-based hypothesis testing framework for differential expression analysis.

Main Results:

• The proposed models provide a richer portrait of differential expression in sequencing data.
• Effectively disentangled sources of transcriptional variation across conditions.
• Successfully summarized and quantified complex transcriptional changes in case-control data.

Conclusions:

• Contrastive latent variable models offer a powerful approach for analyzing RNA-seq data.
• These methods enhance the quantification and summarization of differential gene expression.
• The framework effectively captures complex transcriptional dynamics in biological experiments.