#27 · ML / AI methods

Computational models align and infer across spatial omics and imaging modalities

The dominant thread is computational inference that bridges measurement modalities. An AI pipeline registers NanoSIMS chemical images to electron microscopy ultrastructure, mapping organelle-level distribution of therapeutics within cells. SPOmiAlign uses a feature-matching foundation model to align multimodal spatial omics sections in a modality-agnostic manner, while ST2HE generates virtual H&E images from high-resolution spatial transcriptomics data. The lung digital twin is a separate application of model-based inference: built from the largest clinical ex vivo lung perfusion dataset, it models over 75 parameters and predicts alteplase therapeutic efficacy. What is new is reuse of learned or physics-based models to extract additional information from data already collected.