New CSDE method corrects preprocessing bias in spatial transcriptomics

Researchers have introduced a statistical method called corrected spatial differential expression (CSDE), designed to reduce systematic errors that arise when spatial transcriptomics data are processed before analysis. The work, published in Genome Biology, was led by Pierre Boyeau, Stephen Bates, Can Ergen, Michael I. Jordan and Nir Yosef.

Spatial transcriptomics allows scientists to measure gene activity across tissue sections, but the technique depends on a chain of computational steps. Cells must first be segmented from images, gene expression must be quantified, and individual cells must be assigned to cell types. According to the authors, each of these preprocessing stages is prone to systematic errors, and those errors can distort differential expression analysis, the comparison of gene activity between conditions that underpins much biological discovery.

CSDE builds on an existing statistical framework known as prediction-powered inference. The approach uses a small set of expert-validated cells, in which the preprocessing outputs have been checked by human experts, to correct for the bias introduced during preprocessing. Rather than requiring experts to annotate an entire dataset, the method leverages this limited validation set to produce more reliable differential expression estimates from the full, automatically processed data.

In three case studies reported in the paper, CSDE yielded more reliable differential expression estimates than the prevalent approach, which neglects preprocessing errors altogether. The comparison indicates that accounting for preprocessing bias, even with a modest amount of expert validation, can change the reliability of the resulting gene-level conclusions.

The method addresses a practical bottleneck in large-scale biological assays, where raw outputs such as images and sequences must be converted into structured forms suitable for analysis. The authors note that the published version is shared early as a peer-reviewed, accepted article that remains subject to further editorial edits before the final version of record. As with any new methodological contribution, the reported gains come from the three case studies presented, and the paper’s abstract does not describe broader validation beyond those examples.

Subject of Research: A statistical method that corrects preprocessing bias in spatial transcriptomics differential expression analysis using expert-validated cells.

Article Title: Mitigating bias in spatial transcriptomic pipelines via human feedback with CSDE

Article References: Boyeau, P., Bates, S., Ergen, C., Jordan, M. I., & Yosef, N. (2026). Mitigating bias in spatial transcriptomic pipelines via human feedback with CSDE. Genome Biology. https://doi.org/10.1186/s13059-026-04302-9

Image Credits: AI Generated

DOI: 10.1186/s13059-026-04302-9

Keywords: spatial transcriptomics, differential expression, CSDE, prediction-powered inference, cell segmentation, cell-type annotation, preprocessing bias, Genome Biology, computational biology, expert validation, gene expression, statistical methods