Towards a reliable spatial analysis of missing features via spatially-regularized imputation

Recent spatial transcriptomic (ST) technologies offer a lens for observing the spatial distribution of RNA transcripts in tissues, yet achieving a whole-genome-level spatial landscape remains technically challenging. Multiple computational methods hence have been proposed to impute missing genes from a single-cell reference dataset, while they lack mechanisms of explicitly encoding spatial patterns in the modeling.
    To fill the research gaps, we introduce a computational model, TransImp, that leverages a spatial auto-correlation metric as a regularization for imputing missing features in ST. Evaluation results from multiple platforms demonstrate that TransImp remarkably preserves the spatial patterns, hence substantially improving the accuracy of downstream analysis in detecting spatially highly variable genes and spatial interactions. Therefore, TransImp offers a way towards a reliable spatial analysis of missing features for both matched and unseen modalities, e.g., nascent RNAs.