Figure 1.

Schematic of OverfitDTI

(A) OverfitDTI first performs features learning for drugs and targets; the learned features of drugs X_drug and targets Y_target are concatenated and form the features of the integrated space. Then the concatenated features are fed into DNN to perform the overfitting training. After the model is overfit, the implicit representation f is obtained, which can be used to approximate the nonlinear relationship Z between drugs and targets. For a drug X_i and a target Y_i in this dataset, the nonlinear relationship between them Z_i can be calculated using the implicit representation f, which can then be used to predict the binding scores and determine whether X_i and Y_i interact.

(B) If X_i and Y_i are not in this dataset, a VAE model is used to obtain the features of these unseen data.