FIG. 5.

Tangram's DL framework for aligning/integrating histology and anatomical data with molecular data. Tangram's model for this task is a combination of encoding (using a twin network) and segmentation modules. The twin network learns a similarity metric for brain sections based on anatomical features in images, while the U-Net model is trained to segment five different classes on mouse brain images. This figure was recreated for this manuscript based on illustrations from Biancalani et al.³⁴