The input is the concatenation of the compressed features from all Gene-PCA models, where each feature corresponds to one Gene-PCA. The number of Gene-PCAs is 75,584, so the dimensionality of the input is 75,584 x 1. DiseaseCapsule consists of three layers: a fully connected layer (FC), a primary capsule layer (PrimaryCaps) and a phenotype capsule layer (PhenoCaps). The FC layer consists of 150 neurons followed by ReLU as activation function. The PrimaryCaps is composed of 32 primary capsules. Each of them involves four different convolutional filters (kernel size 5 x 1, stride 2, no padding). PhenoCaps consists of two 16-dimensional vectors. Each phenotype capsule receives input from all 32 primary capsules. The output is a binary classification label (Healthy or ALS).
