Researchers have developed a computational method to map the architecture of human tissues in unprecedented detail. Their approach promises to accelerate studies on organ-scale cellular interactions and could enable powerful new diagnostic strategies for a wide range of diseases. The method, published Oct. 31 in Nature Methods, grew out of the scientists’ frustration with the gap between classical microscopy and modern single-cell molecular analysis. Manually combining single cell data with maps of tissue structure is slow and tedious, though. Machine learning algorithms have shown some potential for automating the process, but they’re limited by the data used to train them. To address that, Kim and his colleagues developed an unsupervised computational strategy, using a combination of single-cell gene expression profiles and cells’ locations to define structural regions within a tissue. The researchers used the new method to generate detailed maps of several types of tissues, identifying and quantifying new aspects of microanatomy — the patterns that emerge at small scale when cells interact and that determine the ultimate function of tissue.