FlashDeconv

FlashDeconv estimates cell type proportions from spatial transcriptomics data with O(N) time complexity using leverage-score importance sampling combined with sparse spatial regularization.

Key Features:

• Ultra-fast deconvolution: Processes 1 million spots in approximately 3 minutes and 1.6 million bins in 153 seconds on commodity CPU hardware without requiring GPU acceleration.
• Structure-preserving randomized sketching: Employs leverage-score importance sampling to reduce computational complexity while preserving data structure.
• Sparse spatial regularization: Incorporates spatial constraints to improve deconvolution accuracy across continuous tissue domains.
• Multi-resolution compatibility: Operates at native resolution without requiring coarsening or sub-sampling, enabling analysis from 8 µm to 64 µm and beyond.
• Continuous proportion output: Preserves mixed spatial niches as continuous cell type proportions rather than discrete labels.

Scientific Applications:

• Tissue-specific resolution horizon identification: Reveals sharp 8-16 µm transitions in mouse intestinal epithelium versus gradual degradation in brain cortex through systematic multi-resolution Visium HD analysis.
• Rare cell niche quantification: Enables sequencing-based quantification of Tuft cell chemosensory niches with 16.2-fold stem cell enrichment at sub-16 µm resolution.
• Tumor microenvironment characterization: Identifies neutrophil inflammatory microdomains co-localized with LAMP3+ mRegDC at the tumor-stroma interface in human colorectal cancer.
• Cross-modality validation support: Provides continuous proportion estimates that can be validated against independent single-cell spatial and other cross-modality data.

Methodology:

FlashDeconv combines leverage-score importance sampling with sparse spatial regularization to estimate cell type proportions from spatial transcriptomics data.