Fig. 1.

The upper part is the branch for drugs, and the lower part is the branch for cell lines. Both are inputs of a fully connected network on the right-hand side. The general work-flow of our model is from left to right. The left-hand side is the input data of one-hot representations for drugs and the feature vectors for cell lines. The black square stands for 1 and empty square stands for 0. In the middle, there are a CNN branch to process the drug inputs and a CNN branch to process cell lines inputs respectively. They take the one-hot representations and feature vectors as input data respectively, and their outputs can be interpreted as the abstract features for drugs and cell lines. The structures of the two convolution neural networks are similar. The right-hand side is a fully connected network that does regression analysis from the IC₅₀ to the abstract features from the two CNNs in the middle part