A model-based factorization method for scRNA data unveils bifurcating transcriptional modules underlying cell fate...

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) generates complex data requiring advanced analytical methods.
• Current manifold-learning techniques offer cell-level trajectory insights but struggle with interpretable biological factors.
• Understanding the genetic basis of cellular state transitions is crucial for developmental biology and disease research.

Purpose of the Study:

• To introduce MGPfact^XMBD, a model-based manifold-learning framework for scRNA-seq data analysis.
• To enable factorization of complex developmental trajectories into independent gene set bifurcation processes.
• To facilitate feature selection for specific biological processes and enhance understanding of cell fate determination.

Main Methods:

• Development of the MGPfact^XMBD framework for model-based manifold learning.
• Application of MGPfact^XMBD to analyze 239 diverse scRNA-seq datasets.
• Benchmarking against established manifold-learning methods using quantitative metrics like branch division accuracy and trajectory topology.

Main Results:

• MGPfact^XMBD demonstrated superior performance in quantitative metrics compared to existing methods.
• The framework successfully identified critical pathways and cell types in microglia development, validated by experimental markers.
• Analysis of tumor-associated CD8⁺ T cells revealed evolutionary trajectories and identified novel subtypes predictive of immune checkpoint inhibitor response.

Conclusions:

• MGPfact^XMBD provides a robust framework for scRNA-seq data analysis, enabling feature selection for biological processes.
• The method offers a more nuanced understanding of cellular trajectories and their underlying determinants.
• MGPfact^XMBD advances the study of cell fate determination and has implications for cancer immunotherapy research.