Consensus scHPF Identifies Cell Type-Specific Drug Responses in Glioma by Integrating Large-Scale scRNA-seq.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) data analysis requires dimensionality reduction for interpretation, but clustering methods struggle with subtle cellular subpopulations.
• Existing methods like single-cell hierarchical Poisson factorization (scHPF) identify gene co-expression but can be sensitive to initialization.
• Complex scRNA-seq studies involve multiple individuals, conditions, and batches, necessitating robust integration methods.

Approach:

• We introduce consensus scHPF, an extension of scHPF that analyzes multiple model initializations to ensure robust gene signature identification.
• Consensus scHPF automatically determines the optimal number of factors, improving the analysis of multi-modal posterior distributions in complex datasets.
• The method facilitates uniform analysis of factors across different individuals and experimental conditions.

Key Points:

• Consensus scHPF identifies the most reliable gene signatures and automatically selects the number of factors for scRNA-seq data.
• It enables effective integration of complex, multi-modal datasets, allowing for consistent analysis across diverse samples and conditions.
• The approach was validated on a large-scale meta-analysis of drug-treated human glioma slice cultures.

Conclusions:

• Consensus scHPF provides a robust framework for analyzing complex scRNA-seq data, improving gene signature discovery and factor determination.
• The method successfully integrated data from multiple patients and drug treatments, revealing cell type-specific responses.
• It identified distinct effects of etoposide and topotecan in glioma, consistent with their protein targets, demonstrating its utility in drug response analysis.