scGPD: single-cell informed gene panel design for targeted spatial transcriptomics.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Selecting informative gene panels is crucial for targeted spatial transcriptomics.
• Existing methods often overlook gene correlations, leading to suboptimal tissue heterogeneity capture.
• Prior knowledge or heuristic rules may not fully represent complex biological systems.

Purpose of the Study:

• To introduce single-cell informed Gene Panel Design (scGPD), a deep learning framework for designing gene panels for spatial profiling.
• To leverage single-cell RNA sequencing (scRNA-seq) data for creating compact and nonredundant gene sets.
• To address limitations in current gene panel selection methods for spatial transcriptomics.

Main Methods:

• scGPD utilizes a deep learning approach with a gene-gene correlation-aware gating mechanism.
• The framework analyzes scRNA-seq data to identify informative and diverse gene sets.
• It aims to eliminate redundancy while maximizing biological information capture.

Main Results:

• scGPD outperforms existing methods in reconstructing transcriptome-wide expression from limited gene panels across diverse scRNA-seq datasets.
• Gene panels designed by scGPD demonstrate robust and competitive cell-type classification performance on spatial transcriptomics data.
• Selected gene panels exhibit clear spatial expression patterns, confirming their relevance for spatial analysis and generalization across modalities.

Conclusions:

• scGPD offers a robust and adaptable solution for designing efficient gene panels for spatial transcriptomics.
• The framework enhances tissue mapping and disease characterization by providing informative gene sets.
• scGPD's flexibility allows adaptation for prioritizing genes related to specific diseases or phenotypes.