Related Experiment Video
Updated: Sep 1, 2025
08:13
Robust Comparison of Protein Levels Across Tissues and Throughout Development Using Standardized Quantitative Western Blotting
Published on: April 9, 2019
14.3K
Telescoping bimodal latent Dirichlet allocation to identify expression QTLs across tissues
Ariel Dh Gewirtz1, F William Townes2, Barbara E Engelhardt2,3
1Lewis-Sigler Institute of Integrative Genomics, Princeton University, Princeton, NJ, USA.
Life Science Alliance
|August 17, 2022
Summary
We developed a new method, telescoping bimodal latent Dirichlet allocation (TBLDA), to analyze gene expression and genotype data. This approach effectively identifies expression quantitative trait loci (eQTLs) in complex datasets with one-to-many sample matching.
Area of Science:
- Genomics
- Systems Biology
- Statistical Genetics
Background:
- Expression quantitative trait loci (eQTLs) link genetic variants to gene expression, crucial for understanding gene regulation.
- Traditional eQTL analyses often ignore complex correlations within gene regulatory networks and linkage disequilibrium.
- Existing multimodal frameworks typically require matched sample sizes between genotype and gene expression data, which is not always feasible.
Purpose of the Study:
- To develop a novel probabilistic framework, telescoping bimodal latent Dirichlet allocation (TBLDA), for analyzing gene expression and genotype data.
- To address the challenge of one-to-many sample matching, common in biological datasets where multiple expression samples are linked to a single individual's genotype.
- To leverage raw count data, avoiding potential biases introduced by normalization procedures.
Main Methods:
- Implemented a telescoping bimodal latent Dirichlet allocation (TBLDA) model to learn shared latent topics across gene expression and genotype data.
- Utilized raw RNA sequencing count data to maintain data integrity.
- Separated ancestral structure into a genotype-specific latent space to isolate shared components.
Main Results:
- The TBLDA model successfully captured meaningful biological signals in both gene expression and genotype data across 10 GTEx tissues.
- Identified 4,645 cis-eQTLs and 995 trans-eQTLs by mapping associations between informative features within the learned topics.
- Demonstrated the model's capability to identify eQTLs in one-to-many matched datasets using raw sequencing counts.
Conclusions:
- The TBLDA framework provides a robust method for eQTL discovery in datasets with nested structures and one-to-many sample matching.
- The model effectively integrates genotype and gene expression data, offering insights into context-specific gene regulation.
- The developed TBLDA model and its code are publicly available, facilitating further research in genetic association studies.
