Sparse deconvolution of cell type medleys in spatial transcriptomics.

Area of Science:

• Spatial transcriptomics
• Computational biology
• Machine learning for genomics

Background:

• Accurate cell type mapping in spatial transcriptomics is crucial for understanding tissue function.
• Current deconvolution models often fail due to assumptions of dataset overlap and neglect of biological constraints like sparsity.
• Technical and biological variations can lead to inaccurate cell-type proportion predictions in existing methods.

Purpose of the Study:

• To introduce Weight-Induced Sparse Regression (WISpR), a novel machine learning algorithm for spatial transcriptomics.
• To develop a method that integrates biological constraints, such as sparsity, for more accurate cell-type distribution prediction.
• To improve the biological coherence and accuracy of cell-type mapping, especially in challenging, unmatched datasets.

Main Methods:

• Developed Weight-Induced Sparse Regression (WISpR), a machine learning algorithm.
• Integrated spot-specific hyperparameters and sparsity-driven modeling into the algorithm.
• Leveraged biology-grounded constraints, including sparsity and cell-type variations.

Main Results:

• WISpR accurately predicts cell-type distributions while preserving biological coherence (spatial and functional consistency).
• The algorithm demonstrates superior performance compared to five alternative methods across ten diverse datasets.
• Successfully predicted cellular landscapes in both normal and cancerous tissues, even with unmatched datasets.

Conclusions:

• WISpR provides biologically informed, high-resolution cellular maps by leveraging sparse cell-type arrangements.
• The method offers practical utility for spatial transcriptomics, particularly in settings with noise, sparsity, or reference mismatches.
• WISpR enhances the decoding of tissue organization in both healthy and diseased states.