INSPIRE
We develop INSPIRE, a deep learning-based method for integrating and interpreting multiple spatial transcriptomics (ST) datasets from diverse sources. It integrates information across sections in a shared latent space, where meaningful biological variations from the input sections are preserved, while complex unwanted variations are eliminated. Utilizing this shared latent space, INSPIRE achieves an integrated NMF on gene expressions across sections, decomposing biological signals in different sections into consistent and interpretable spatial factors with associated gene programs. These inferred spatial factors often correspond to distinct cell populations and biological processes within the analyzed tissues.
INSPIRE takes gene expression count matrices and spatial coordinates from multiple ST sections as input, and generates three key outputs: latent representations of cells or spatial spots, non-negative spatial factors for cells or spatial spots, and non-negative gene loadings shared among datasets.
By integrating multiple ST datasets with INSPIRE, users can:
Identify spatial trajectories and major spatial regions consistently across datasets using latent representations of cells or spatial spots.
Reveal detailed tissue architectures, spatial distributions of cell types, and organizations of biological processes in tissues across sections using non-negative spatial factors for cells or spatial spots.
Detect spatially variable genes, identify gene programs associated with specific spatial architectures in tissues, and conduct pathway enrichment analysis using non-negative gene loadings.
Quick Start and Usage Instructions
INSPIRE Tutorials
- Tutorials
- Example 1: Analysis of human dorsolateral prefrontal cortex (DLPFC) dataset.
- Example 2: Analysis of mouse brain slices with complementary views.
- Example 3: Analysis of mouse brain slices across distinct ST technologies.
- Example 4: Analysis of mouse whole-embryo slices across distinct ST technologies.
- Example 5: Analysis of mouse whole-embryo slices across developmental stages.
- Run INSPIRE on whole-embryo slices across multiple developmental stages
- Changes in proportions of spots covered by the identified organs and tissues across stages
- Analysis of the liver region identified by INSPIRE
- Analysis of INSPIRE’s identified brain region III
- Analysis of the spatial factor related to choroid plexus and its associated gene signature
- Analysis of the spatial factor related to smooth muscle cells and its associated gene signature
- Example 6: Spatial registration between two MERFISH hypothalamus sections.
- Example 7: Spatial registration between two STARmap PLUS hippocampus sections.
- Example 8: 3D reconstruction of mouse embryo by integrating multiple 2D slices.
- Example 9: Analysis of mouse skin data undergoing wound healing
- Example 10: Analysis of human breast cancer Xenium data, revealing microenvironment heterogeneity across tumor subtypes