Platform-specific count-matrix preprocessing workflows for spatial transcriptomics data analysis

Spatial transcriptomics (ST) data analysis relies heavily on preprocessing of count matrices to eliminate technical variation while retaining biological signals. However, the optimal strategies for preprocessing vary greatly across different platforms. To address this, researchers systematically evaluated 371 preprocessing workflows across 45 datasets representing 11 mainstream ST platforms, using specific metrics to assess performance.

They also examined image preprocessing steps, such as cell segmentation and post-segmentation transcript processing, on representative imaging-based ST platforms. The findings revealed that there is no one-size-fits-all count-matrix preprocessing workflow for all platforms. Moreover, data from the same platforms showed consistent optimal preprocessing workflows.

Key data structure characteristics and preprocessing steps were identified as crucial factors influencing the performance of count-matrix preprocessing. The researchers developed platform-specific and data structure-guided recommendations to help users choose the best count-matrix preprocessing workflow for their specific needs.

These recommendations were validated across a wider range of downstream tasks, demonstrating that the data-driven approach uncovers subtle biologically meaningful spatial patterns in new datasets that might be missed with suboptimal preprocessing methods. Overall, these findings provide best practices for ST data count-matrix preprocessing.