Model-based prediction of spatial gene expression via generative linear mapping.

Area of Science:

• Computational Biology
• Genomics
• Developmental Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding multicellular systems.
• Existing computational methods for spatial transcriptomics lack accuracy and biological interpretability.
• Model-free frameworks limit the biological insights derived from scRNA-seq data.

Purpose of the Study:

• Introduce Perler, a novel model-based method for integrating scRNA-seq and in situ hybridization (ISH) data.
• Develop a biologically interpretable model to calibrate differences between scRNA-seq and ISH datasets.
• Achieve accurate reconstruction of spatial transcriptomes with enhanced biological interpretability.

Main Methods:

• Developed a generative linear mapping model based on a Gaussian mixture model (GMM) with the Expectation-Maximization (EM) algorithm.
• Integrated scRNA-seq data with reference ISH data using the Perler model.
• Validated the method on diverse biological systems including Drosophila embryos, zebrafish embryos, mammalian liver, and mouse visual cortex.

Main Results:

• Perler accurately predicts spatial gene expression from scRNA-seq data across multiple species and tissues.
• The reconstructed transcriptomes showed no over-fitting to the ISH data.
• Perler preserved the temporal information present in the scRNA-seq data.

Conclusions:

• Perler provides a generalizable and biologically interpretable framework for spatial transcriptome reconstruction.
• The model-based approach overcomes limitations of existing model-free methods.
• Enables accurate spatial transcriptome reconstruction for any multicellular system, advancing biological discovery.