Deep Personalized Prediction of MPA-AUC with Attentive Interactions in Kidney Transplantation

It is important to monitor the plasma concentration degree in kidney transplantation and adjust drug dosages accordingly. The selection of plasma concentration sampling time points is a challenging task and the limited sampling strategy is used traditionally. The existing solutions have not considered characteristics of individual patients and had room for improvement. In this work, we propose a new deep learning model, Deep Personalized Prediction of Mycophenolic Acid Area Under Curve with Attentive Interactions (DPMPA-AUC) to select the plasma concentration sampling time points. This model has the capability to take account of the factors of individual patients. We use the soft attention method to select the sampling time points for the plasma concentration. We analyze the correlation between plasma concentration degree and the Mycophenolic Acid density distribution after kidney transplantation. We consider 2 to 4 sampling time points in this work. We use individual factors as the attention weights of the plasma concentration degree. We develop a Long Short Term Memory Dilated Convolutional Neural Networks with the soft attention mechanism to predict Mycophenolic Acid density during 12 hours after taking the drug. The experimental results show that our solution has better performance comparing with existing solutions.