Single-cell gene regulatory network prediction by explainable AI.

Area of Science:

• Computational Biology
• Genomics
• Cancer Research

Background:

• Cancer's molecular heterogeneity fuels treatment resistance and relapse.
• Single-cell sequencing provides descriptive insights but lacks functional understanding of gene regulation.
• Existing methods often predict average networks, missing individual cell variations.

Purpose of the Study:

• To develop scGeneRAI, an explainable deep learning model for inferring single-cell gene regulatory networks (GRNs).
• To analyze functional gene regulatory patterns in individual cells from static single-cell RNA sequencing data.
• To characterize tumor cell-specific regulatory subnetworks and their role in cancer heterogeneity.

Main Methods:

• Proposed scGeneRAI, an explainable deep learning approach utilizing layer-wise relevance propagation (LRP).
• Inferred GRNs from static single-cell RNA sequencing (scRNA-seq) data at the single-cell level.
• Benchmarked scGeneRAI using synthetic data and applied it to human lung cancer scRNA-seq data.

Main Results:

• scGeneRAI successfully inferred single-cell GRNs, distinguishing tumor from normal cells.
• Identified characteristic network patterns specific to tumor cells and subgroups of patients.
• Revealed subnetworks unique to specific tumor cell populations, highlighting regulatory heterogeneity.

Conclusions:

• scGeneRAI enables functional insights into gene regulation at the single-cell level.
• The method effectively characterizes molecular heterogeneity driving cancer.
• Facilitates deeper understanding of gene regulatory differences within and across tumors for personalized medicine.