scRegulate: Single-Cell Regulatory-Embedded Variational Inference of Transcription Factor Activity from Gene...

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Accurate inference of transcription factor (TF) activity from single-cell RNA sequencing (scRNA-seq) is crucial but challenging.
• Existing methods often rely on static databases and deterministic assumptions, limiting their ability to capture dynamic regulatory interactions.

Purpose of the Study:

• To develop a novel deep learning framework, scRegulate, for inferring TF activities and gene regulatory networks (GRNs) from scRNA-seq data.
• To integrate prior biological knowledge with data-driven inference for improved accuracy and interpretability.

Main Methods:

• scRegulate employs a generative deep learning framework with variational inference.
• It incorporates gene regulatory network (GRN) priors and structured biological constraints within a probabilistic latent space model.
• The framework was benchmarked on synthetic and experimental datasets, including Perturb-seq and PBMC scRNA-seq data.

Main Results:

• scRegulate demonstrated superior performance in inferring TF activities and GRNs compared to existing methods on synthetic datasets (AUROC 0.71-0.86, AUPRC 0.80-0.95).
• It accurately recapitulated TF knockdown effects in experimental data and identified key TFs like ELK1, EGR1, and CREB1.
• Application to PBMC data revealed cell-type-specific GRNs and enabled accurate cell type clustering through TF embeddings.

Conclusions:

• scRegulate provides a scalable, interpretable, and powerful framework for TF activity and GRN inference from scRNA-seq data.
• The method effectively integrates prior biological knowledge with data-driven approaches.
• scRegulate advances the analysis of transcriptional regulation and cellular heterogeneity.