New algorithm makes maps of gene activity easier to compare while preserving cell-level detail

Researchers at Kanazawa University and Sapienza University of Rome have developed a computational method called Domain Elastic Transform (DET) that aligns gene activity maps from tissue samples while preserving cell-level detail. Spatial transcriptomics can reveal the activity of thousands of genes across a tissue at single-cell resolution, but comparing these maps is challenging due to variations in tissue shape and orientation.

DET addresses this issue by directly aligning the individual measurement locations and their gene-activity values, without converting the measurements into a regular grid of pixels. This preserves the fine structures and detailed cell-level information present in the original data. The method works by estimating likely matches between cells in two maps based on their positions and gene activity similarities, then adjusting cell positions to gradually refine the alignment.

DET does not require pre-aligned training examples or manually identified matches, making it training-free and unsupervised. In tests using mouse brain maps rotated and shifted to simulate natural tissue variations, DET outperformed other methods in overlapping tissue maps, maintaining neighboring cell relationships, and preserving gene-activity patterns.

The researchers also tested DET on developing mouse embryo maps from two stages, demonstrating its ability to align maps of different shapes and cell compositions. The method requires relatively little memory usage, even when handling large datasets with up to 1 million measurement points per tissue slice.