cellsig plug-in enhances CIBERSORTx signature selection for multidataset transcriptomes with sparse multilevel
Md Abdullah Al Kamran Khan1, Jian Wu2, Yuhan Sun1
1Department of Microbiology and Immunology, The University of Melbourne at The Peter Doherty Institute for Infection and Immunity, Parkville, VIC 3010, Australia.
Bioinformatics (Oxford, England)
|November 12, 2023
Summary
CellSig, a new Bayesian model, improves cell-type signature estimation from bulk RNA sequencing data by handling complex data structures and non-overlapping gene sets. This enhances marker gene identification for various biological applications.
Area of Science:
- Genomics
- Bioinformatics
- Computational Biology
Background:
- Precise cell-type transcriptome characterization is crucial for understanding cellular dynamics and clinical applications.
- Single-cell RNA sequencing has advanced cell-type profiling, but challenges remain with data heterogeneity and non-overlapping gene sets across studies.
- Existing tools like CIBERSORTx struggle with hierarchical data and non-overlapping gene sets, often requiring data imputation or filtering.
Purpose of the Study:
- To develop a novel Bayesian sparse multilevel model, named cellsig, to enhance cell-type signature estimation.
- To address limitations in current methods regarding data heterogeneity, hierarchical structures, and non-overlapping gene sets in bulk and pseudobulk RNA sequencing data.
- To improve the accuracy and robustness of cell-type marker gene identification.
Main Methods:
- Developed a Bayesian sparse multilevel model (cellsig) accounting for multilevel effects and gene-set sparsity.
- Applied cellsig to a curated Human Bulk Cell-type Catalogue comprising 1435 samples across 58 datasets.
- Evaluated cellsig's performance in cell-type marker gene ranking compared to existing approaches.
Main Results:
- CellSig significantly improves cell-type marker gene ranking performance.
- The model effectively handles large-scale, heterogeneous bulk and pseudobulk RNA sequencing data with non-overlapping gene sets.
- Demonstrated enhanced performance on a harmonized dataset of 1435 samples.
Conclusions:
- CellSig offers a robust approach for cell-type signature selection from complex transcriptomic data.
- The method has significant implications for marker gene validation, single-cell annotation, and deconvolution benchmarks.
- Provides a valuable tool for researchers working with large-scale RNA sequencing datasets.
