Extended Data Fig. 2: The pipeline to test the deep neural network model’s generalizability and accuracy.
From: TransBrain: a computational framework for translating brain-wide phenotypes between humans and mice

(a) One of the six AHBA donors was excluded as the independent test set, while the remaining five were used for training. Each AHBA/single-nucleus donor pair underwent independent integration. For all combinations, integration was repeated 100 times, and each dataset underwent 10 rounds of model training and testing. During training, stratified 10% validation sets were defined based on pre-integration AHBA sample labels per brain region.