Synthetic control removes spurious discoveries from double dipping in single-cell and spatial transcriptomics data...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell and spatial transcriptomics analysis faces a 'double dipping' pitfall where clustering and differential expression (DE) gene identification use the same data.
• This can lead to false-positive markers, especially with spurious clusters, complicating cell type or spatial domain definitions.

Purpose of the Study:

• To introduce ClusterDE, a statistical method to reliably identify DE gene markers post-clustering in transcriptomics data.
• To control the false discovery rate (FDR) irrespective of clustering algorithm performance.

Main Methods:

• ClusterDE employs synthetic null data as an in silico negative control to detect and eliminate double dipping effects.
• The method generates a single cell type or spatial domain null distribution for robust statistical testing.

Main Results:

• ClusterDE effectively controls the FDR, identifying genuine cell-type and spatial-domain markers while distinguishing them from housekeeping genes.
• The method's output can inform decisions on merging ambiguous clusters based on marker discovery.

Conclusions:

• ClusterDE offers a robust solution to the double dipping problem in transcriptomics data analysis.
• The method is compatible with widely used pipelines like Seurat and Scanpy, enhancing its practical applicability.