Fig. 1

The autoencoder model of neurotypical controls and individuals with autism. A A schematic of the autoencoder model consisting of four encoder, one bottleneck, and four decoder layers. The diffusion tractography-based structural connectivity (SC) is entered into the autoencoder. B The upper panels show the loss graphs of training and validation datasets according to epochs for neurotypical controls (left) and individuals with autism (right). The scatter plots on the bottom represent correlations between the actual and reconstructed structural connectivity of the test data