Machine intelligence design of 2019-nCoV drugs

Kaifu Gao; Duc Duy Nguyen; Rui Wang; Guo-Wei Wei

properties. The next component is the MathPose model which is used to predict the three-dimensional (3D) structure information of the compounds selected by 2DFP-DNN. The bioactivities of those compounds are further estimated by the structure-based deep learning model named MathDL. 18 The druggable properties predicted by this last component of our GNC are used as an indicator to select the promising drug candidates. Figure 1 : A schematic illustration of the generative network complex. SMILES strings (SSs) are encoded into latent space vectors via a gated recurrent neural network (GRU)-based encoder. These vectors are modified in the molecule generator to achieve desirable druggable properties, such as binding affinity, partition coefficient (LogP), similarity, etc., predicted by pre-trained deep neural networks (DNNs). The resulting drug-like molecules are translated into SSs by a GRU-based decoder. The physical properties of these SSs are validated by 2D fingerprint-based multitask DNNs. Promising drug candidates are fed into a MathPose unit to generate 3D structures, which are further validated by a mathematical deep learning (MathDL) center to select new drug candidates. 18