SuperFocus enables whole-slide cell-level spatial omics from spot-based measurements and histology across modalities

Spatial omics data acquisition is constrained by an inherent trilemma: balancing experimental cost, spatial resolution, and profiling scope (in terms of both plexity and field of view). Currently, spot-based technologies coupled with histological images represent one of the most scalable solutions, yet they lack single-cell precision and are often limited in field of view for omics. Here, we introduce SuperFocus, an AI-powered framework that predicts omics features with high fidelity for individual cells across whole-slide images from spotlevel omics data acquired on the same or adjacent slides. It sub-stantially relaxes the trilemma by achieving single-cell resolution and whole-slide field of view without increasing experimental cost or sacrificing plexity. SuperFocus implements a novel constrained cascading imputation scheme to minimize hallucination and outputs two per-feature quality-control scores that flag potentially unreliable predictions for each feature. When predicting for cells outside the profiled field of view, SuperFocus additionally provides a prediction-credibility score for each cell that prevents overconfidence when relevant histological patterns are undersampled or absent from the training data. We validate the fidelity of SuperFocus by benchmarking against multiple ground-truth single-cell and subcellular resolution spatial datasets across different platforms, demonstrating highly concordant cell-level feature recovery. We evaluate SuperFocus on both spatial transcriptomics and spatial epigenomics modalities and observe consistent performance over diverse platforms, tissues, and conditions. We demonstrate its utility by resolving the MALT lymphoma microenvironment at single-cell resolution using spot-level Patho-DBiT data, and by performing single-cell resolution motif analysis of the human hippocampus using spot-level spatial ATAC-seq. Overall, SuperFocus provides an effective in silico solution to the spatial omics trilemma, empowering single-cell-resolved spatial analysis from standard spot-level inputs and histological images.