Evaluation of a deep generative computational framework under constrained conditions for multi-regional placental...

Area of Science:

• Genomics
• Computational Biology
• Reproductive Medicine

Background:

• The placenta's heterogeneity is key to preeclampsia (PE), but high single-cell RNA sequencing (scRNA-seq) costs limit clinical studies.
• scSemiProfiler infers single-cell transcriptomes from bulk RNA-seq but traditionally uses active learning for sample selection.
• Evaluating a simplified scSemiProfiler workflow without active learning in complex placental tissues is crucial for cost-effective analysis.

Purpose of the Study:

• To assess the performance of a simplified scSemiProfiler workflow that omits active learning for analyzing placental transcriptomes.
• To determine the workflow's effectiveness in capturing cell composition, marker genes, and functional pathways in multi-regional placental tissues.
• To evaluate the impact of gene-count filtering on data quality and analysis outcomes.

Main Methods:

• Generated 14 scRNA-seq libraries from basal plate, villi, and chorioamniotic membranes from 5 women (1 control, 4 PE).
• Integrated scRNA-seq data with 118 bulk RNA-seq datasets using scSemiProfiler without active learning.
• Compared semi-profiled and real-profiled data for cell composition, marker genes, Gene Ontology (GO) enrichment, and assessed gene-count filtering effects.

Main Results:

• Semi-profiled data largely recapitulated tissue-specific cell composition, though rare populations showed deviations.
• Canonical marker genes remained identifiable despite reduced gene expression intensity and detection rates.
• Filtering cells with <10 detected genes improved concordance and visualization with minimal impact on GO consistency.
• The method captured pathway-level functional variation and disease-associated GO changes but was less reliable for rare cell types and subtle gene differences.

Conclusions:

• The simplified scSemiProfiler workflow, without active learning, effectively preserves broad placental transcriptional patterns.
• The workflow demonstrates limitations in accurately profiling rare cell types and detecting subtle disease-related gene expression contrasts.
• Further refinement may be needed for high-resolution analysis of complex placental tissues and disease states.