Identification of transcriptional programs using dense vector representations defined by mutual information with...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cell phenotypes through transcriptional processes.
• Current dimensionality reduction methods often aggregate sparse gene data, neglecting inter-gene relationships.
• This aggregation can lead to a loss of critical biological information and hinder accurate cell type classification.

Purpose of the Study:

• To introduce GeneVector, a novel scalable framework for dimensionality reduction in scRNA-seq data.
• To demonstrate GeneVector's ability to model gene co-expression and overcome data sparsity.
• To showcase GeneVector's utility in identifying transcriptional programs and classifying cell types.

Main Methods:

• GeneVector employs a vector space model utilizing mutual information to capture gene co-expression.
• It performs dimensionality reduction with respect to gene co-expression patterns.
• Latent space arithmetic in a lower-dimensional gene embedding is used for analysis.

Main Results:

• GeneVector successfully captured phenotype-specific pathways across four scRNA-seq datasets.
• The framework demonstrated effectiveness in batch effect correction for scRNA-seq data.
• Interactive cell type annotation and identification of pathway variations were achieved.

Conclusions:

• GeneVector offers a powerful approach to dimensionality reduction in scRNA-seq by modeling gene co-expression.
• It provides a scalable and effective tool for cell type classification and pathway analysis.
• GeneVector enhances the interpretation of complex single-cell transcriptional data.