Fig. 1
From: iMAP: integration of multiple single-cell datasets by adversarial paired transfer networks
Overview of the iMAP algorithm. a The expression profiles modeled by a function of two independent factors: biological variations and measuring processes. b Three feed-forward neural networks deployed for the first stage of iMAP. c Information flows and two losses functions used in the first stage. Lr, reconstruction loss; Lc, content loss. d A within-batch random walk-based procedure adopted to extend the MNN pairs. These extended MNN pairs are called rwMNN pairs. e A GAN structure used to remove the batch effects based on the rwMNN pairs