A pharmacokinetic-pharmacodynamic model based on the SSA-1DCNN-Attention network and the semicompartment method

To solve the problem of inaccurate prediction caused by the lack of representativeness of samples due to the small sample size of the collected clinical data when using machine learning methods to predict drug concentration in plasma and describe the hysteresis phenomenon of drug effect lagging behind plasma drug concentration, this paper proposes a pharmacokinetic-pharmacodynamic (PK-PD) model based on the SSA-1DCNN-Attention network and the semicompartment method. First, a one-dimensional convolutional neural network (1DCNN) is established, and the attention mechanism is introduced to determine the importance of each physiological and biochemical parameter. The sparrow search algorithm (SSA) is used to optimize the parameters of the network to improve the prediction accuracy after data enhancement through the synthetic minority oversampling technique (SMOTE) method. After constructing the time-concentration relationship of the drug through the SSA-1DCNN-Attention network, the concentration-effect relationship of the drug is established by using the semicompartment method to synchronize the drug effect with the concentration. At last, the phenobarbital (PHB) combined with Cynanchum otophyllum saponins to treat epilepsy was taken as an example to validate the proposed method.