2.1. Predictive modeling by deep learning

A deep learning algorithm was used to develop the predictive model. To make this model, deep learning online server was used (http://deepscreening.xielab.net) (Liu et al., 2019). The CHEMBL dataset (CHEMBL3927) was used, which contained IC50 values for the inhibition activity of the SARS coronavirus 3C-like proteinase. This CHEMBL dataset was preprocessed for molecular vectorization by applying PubChem fingerprint which generates 881 fingerprints using PaDEL software (Yap, 2011). The PubChem fingerprints were used to construct a regression model by applying deep recurrent neural networks (RNN). Several models were generated by manual optimization of hyperparameters like learning rate, epoch, batch size, number of neurons, hidden layers, etc to select the best model (Table 1). All the hidden layers used ReLU activation function (y = max (0, 1)), while the output layer used a sigmoid function.

Manual optimization of hyperparameters to select the best deep learning model.

The bold one is the best regression model in terms of R², MSE, RMSE, MAE, and Loss.