Gene selection for optimal prediction of cell position in tissues from single-cell transcriptomics data

Area of Science:

• Developmental Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNAseq) provides high-resolution gene expression data but typically loses spatial information.
• Spatial transcriptomics methods preserve cell location but often have limited throughput and gene coverage.
• Integrating scRNAseq with spatial gene expression data is crucial for comprehensive tissue analysis.

Purpose of the Study:

• To benchmark computational methods for reconstructing cell spatial organization from scRNAseq data.
• To identify optimal gene sets for predicting cell locations in developing tissues.
• To evaluate the generalizability of these methods across different species.

Main Methods:

• The DREAM Single-Cell Transcriptomics challenge utilized Drosophila embryo scRNAseq data.
• A reference atlas of in situ hybridization data served as the 'silver standard' for spatial validation.
• Participating teams developed and compared diverse algorithms for gene selection and spatial prediction.

Main Results:

• Methods successfully localized clusters of cells, highlighting the importance of predictor gene selection.
• Effective predictor genes exhibited high expression entropy, spatial clustering, and included key developmental genes (e.g., gap, pair-rule genes).
• Top methods applied to zebrafish embryo data showed comparable performance and gene properties, indicating generalizability.

Conclusions:

• The challenge successfully benchmarked spatial reconstruction methods for scRNAseq data.
• Identifying informative genes is critical for accurately mapping cells to their spatial locations.
• The developed methods and identified gene properties are generalizable for reconstructing tissue organization in developmental contexts.