Differential expression analysis for spatially correlated data using smiDE.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Imaging spatial transcriptomics enables analysis of cell state responses within microenvironments.
• Standard differential expression methods face challenges with spatial data, including segmentation errors and spatial correlation.
• Ignoring these spatial data issues can lead to a high rate of false positive findings in differential expression analysis.

Purpose of the Study:

• To address fundamental challenges in differential gene expression analysis for imaging spatial transcriptomics data.
• To develop robust methods that account for segmentation errors and cell-cell correlations.
• To provide an accessible R package implementation for these solutions.

Main Methods:

• Development of statistical methods to correct for bias introduced by segmentation errors in spatial transcriptomics.
• Implementation of models that account for the spatial correlation between neighboring cells.
• Integration of these solutions into a user-friendly R package named smiDE.

Main Results:

• Demonstration that ignoring segmentation errors and spatial correlation inflates statistical significance and leads to numerous false discoveries.
• Validation of the proposed methods in correcting for these biases.
• Successful implementation of the corrected methods within the smiDE R package.

Conclusions:

• Accurate differential gene expression analysis in spatial transcriptomics requires addressing segmentation errors and spatial correlation.
• The smiDE R package provides a robust framework for reliable differential expression analysis in imaging spatial transcriptomics.
• The developed methods mitigate false discoveries, improving the accuracy of biological insights from spatial transcriptomics data.