Sparse deconvolution of cell type medleys in spatial transcriptomics

Abstract

Mapping cell distributions across spatial locations with whole-genome coverage is essential for understanding cellular responses and signaling pathways. However, current deconvolution models often assume strong overlap between reference and spatial datasets, neglecting biological constraints like sparsity and cell-type variations. As a result, these methods rely on brute-force algorithms that ignore tissue complexity, leading to inaccurate predictions, particularly in heterogeneous or unmatched datasets. We introduce Weight-Induced Sparse Regression (WISpR), a machine learning algorithm that integrates spot-specific hyperparameters and sparsity-driven modeling. Unlike brute-force methods, WISpR accurately predicts cell-type distributions while maintaining biological coherence, even in unmatched datasets. Benchmarking against five leading methods across ten datasets, WISpR consistently outperformed competitors and predicted cellular landscapes in both normal and cancerous tissues. By leveraging sparse cell-type arrangements, WISpR provides biologically informed, high-resolution cellular maps. Its ability to decode tissue organization in both healthy and diseased states marks a major advancement in spatial transcriptomics, setting a new standard for accurate deconvolution.