Integrating Graph Convolutional Networks for Missing Gene Expression Imputation.

Area of Science:

• Biomedical Sciences
• Genomics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) provides cellular resolution but lacks spatial context.
• Spatial transcriptomics offers gene expression with spatial mapping but has limited gene throughput.
• Accurate spatial gene expression prediction is crucial for understanding cellular function in situ.

Purpose of the Study:

• To develop a novel computational method, GCNgene, for predicting spatial gene distribution.
• To integrate scRNA-seq and spatial transcriptomics data for enhanced spatial transcriptomic analysis.
• To enable whole-transcriptome-level spatial gene expression profiling.

Main Methods:

• GCNgene utilizes a graph convolutional network (GCN) to embed spatial transcriptomics data.
• A learned rule reconstructs gene expression by combining reference scRNA-seq data and cell-type proportions.
• The method integrates spatial and single-cell datasets for accurate gene expression prediction.

Main Results:

• GCNgene accurately predicts the spatial distribution of undetected RNA transcripts.
• The approach enables the reconstruction of gene expression levels in spatial contexts.
• Successful integration of diverse transcriptomic datasets for spatial analysis.

Conclusions:

• GCNgene offers a powerful computational solution for predicting spatial gene expression.
• This method addresses the limitations of current spatial transcriptomic technologies.
• Facilitates a deeper understanding of cellular heterogeneity and tissue architecture through spatial gene profiling.