FlashDeconv enables atlas-scale, multi-resolution spatial deconvolution via structure-preserving sketching

What's Changed New Features Add rho_sparsity parameter to deconvolve function for L1 sparsity penalty control Add preprocess option to deconvolve function Add reproducibility support with random_state parameter Add Stereo-seq platform guide Bug Fixes Fix normalization for sparse matrices in deconv Handle empty inputs and validate graph coordinates Handle sparse references and clamp k in KNN graph Clamp k bounds and use clamped min_k in estimator Correct Visium HD typical UMI/spot values based on published data Improvements Input validation and robustness fixes across modules Validation, dtype fixes, and safe correlation handling Performance and robustness improvements Change lambda_spatial default to auto Change license from GPL-3.0 to BSD-3-Clause Documentation Add comprehensive spatial deconvolution tutorial notebook Add reference data quality guide Clarify and standardize rho_sparsity scaling and docs Document the impact of unknown cells in deconvolution Revise and expand README with updated features and benchmarks Add Zenodo DOI badge to README Full Changelog: https://github.com/cafferychen777/FlashDeconv/compare/v0.1.3...v0.1.6