Fig. 1.

The network architecture of DOVE. DOVE takes atom positions and potentials in a 20*20*20 input cube that is placed at the docking interface of a decoy and predicts if the decoy is in the CAPRI acceptable quality or not. The number of filters in each layer were 100, 200, 200, 400, 400; 20, 18, 16, 8, 6, 3 are the output cube size of each layer; 10 800, 1000, 100 denotes the number of neurons for FC layer. Block means 3D convolutional operation block that outputs data in 3D cube; Flat is to make a 1D vector from a 3D cube; Pool is a max-pooling operation and FC is fully-connected network. Dropout of 0.3 was applied to FC