Accurate and well-powered case–control analysis of spatial molecular data

As spatial molecular data grow in scope, there is a pressing need to identify disease-associated spatial structures. Current approaches typically make restrictive assumptions such as representing tissue regions by abundances of discrete cell types and samples by abundances of discrete niches; this risks overlooking important signals. Here we introduce variational inference-based microniche analysis (VIMA), a method combining deep learning with principled statistics to discover disease-associated spatial features with greater flexibility and precision. VIMA trains an ensemble of variational autoencoders to summarize the contents of every small tissue patch in a dataset via numeric ‘fingerprints’. It uses these to define many data-dependent, overlapping ‘microniches’ and meta-analyzes them to identify microniches whose abundance correlates significantly with case−control status. We confirm VIMA’s calibration, power and spatial accuracy in simulations. We then apply VIMA to spatial datasets spanning three diseases and spatial modalities, recapitulating known biology and identifying new spatial features of disease. VIMA uses deep-learning architecture to identify differentially enriched features within spatial datasets.