INSPIRE: interpretable, flexible and spatially-aware integration of multiple spatial transcriptomics datasets from...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics technologies generate diverse datasets for exploring tissue organization and function.
• Integrating data from different samples, technologies, and developmental stages presents a significant challenge.

Purpose of the Study:

• To present INSPIRE, a deep learning method for the integrative analysis of multiple spatial transcriptomics datasets.
• To enable spatially informed and adaptable integration of data from varying sources.

Main Methods:

• INSPIRE utilizes graph neural networks and adversarial learning for data integration.
• Non-negative matrix factorization is incorporated to uncover interpretable spatial factors and gene programs.
• The method is applied to diverse datasets including human cortex, mouse brain, and embryonic development atlases.

Main Results:

• INSPIRE effectively integrates data from distinct profiling technologies, borrowing information across sources.
• The method identifies detailed biological signals, revealing tissue architectures, cell type distributions, and biological processes.
• INSPIRE demonstrates superior performance in elucidating dynamical changes during embryonic development.

Conclusions:

• INSPIRE provides a powerful and versatile tool for the integrative analysis of spatial transcriptomics data.
• The method facilitates the construction of 3D tissue and organismal models from multiple slices.
• INSPIRE advances the understanding of tissue organization, cell function, and developmental dynamics.