DEEPsc: A Deep Learning-Based Map Connecting Single-Cell Transcriptomics and Spatial Imaging Data.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides cellular resolution but often loses spatial context.
• Computational methods exist to infer spatial information using reference atlases, but lack comprehensive performance analysis.
• Existing methods may lack generalizability across different biological systems.

Purpose of the Study:

• To develop and evaluate a novel system-adaptive deep learning method (DEEPsc) for imputing spatial information onto scRNA-seq data.
• To introduce a comprehensive suite of metrics for evaluating spatial mapping accuracy, precision, robustness, and generalizability.
• To compare DEEPsc against existing methods across diverse biological systems.

Main Methods:

• Development of DEEPsc, a deep learning-based approach for spatial imputation.
• Creation of a standardized evaluation framework with novel metrics for spatial mapping.
• Comparative analysis of DEEPsc and four other methods on four distinct biological datasets.

Main Results:

• DEEPsc demonstrates comparable accuracy to existing methods for spatial imputation.
• DEEPsc achieves an improved balance between precision and robustness in spatial mapping.
• The study provides a comprehensive evaluation of multiple spatial imputation techniques.

Conclusions:

• DEEPsc offers a data-adaptive tool for integrating scRNA-seq and spatial imaging data.
• The developed metrics enable rigorous assessment of spatial mapping method performance.
• An open-source software implementation with a uniform API facilitates spatial exploration of cell fate decisions in scRNA-seq data.