SpaHDmap: interpretable high-definition embedding mapping

SpaHDmap is a multi-modal neural network that takes advantage of the high-dimensionality of spatial transcriptomics data and the high-definition of image data to achieve interpretable high-definition dimension reduction. The high-dimensional expression data enable refined functional annotations and the high-definition image data help to enhance the spatial resolution.

Based on the high-definition embeddings and the reconstruction of gene expressions, SpaHDmap can then perform high-definition downstream analyses, such as spatial domain detection, gene expression recovery, and identification of embedding-associated genes as well as high-definition cluster-associated genes.

Key Features of SpaHDmap

• Integrates deep learning with NMF for spatial transcriptomics analysis

• Supports various spatial transcriptomics platforms (e.g., 10X Visium, Stereo-seq)

• Generates interpretable high-definition embeddings and spatial clusters

• Enables simultaneous analysis of multiple samples

• Provides visualization tools for easy interpretation of results

• Offers both Python API and command-line interface for flexibility

• Allows selection of the optimal embedding dimension for improved interpretability.

• Supports transferring a pre-trained model to new datasets.

• Enables spatial domain refinement to capture biological sub-structures.

• Facilitates differential expression and GO enrichment analysis for detected high-resolution clusters.

Getting started with SpaHDmap

To quickly get started with SpaHDmap, please refer to the Installation and Tutorials. For more details, please refer to our paper.