Prediction Of Mycophenolic Acid Exposure After Kidney Transplantation Based On Few-shot Learning

In clinical practice,patients usually need to take Mycophenolic Acid(MPA)and other immunosuppressant drugs after kidney transplantation to inhibit the anti-rejection reaction of organ transplantation and reduce the antibody immune response.Clinical studies have found that there are great individual differences in the adverse reactions of kidney transplantation patients after taking MPA drugs,which is closely related to the exposure level of MPA drugs in the patients.Mycophenolic acid exposure level after renal transplantation refers to the Area under the AUC(Area under curve)of plasma drug concentration-time curve,i.e.,MPA-AUC0-12h.It reflects the exposure of mycophenolic acid in kidney transplantation patients.Too high MPA-AUC0-12h will lead to increased acute rejection and too low MPA-AUC0-12h will lead to increased incidence of adverse reactions.Therefore,it is of great clinical significance to accurately calculate the exposure level of mycophenolic acid in patients after kidney transplantation.The usual method to calculate mycophenolic acid exposure level in medical clinic is to take at least 9-12 plasma samples within 12 hours after patients taking MPA drugs,then draw blood drug concentration-time curve based on MPA to calculate the mycophenolic acid exposure level(i.e.,the area under the curve of MPA concentration).In this process,medical staff need to collect blood regularly at 9-12 time points,which is a huge workload.Meanwhile,multiple blood collection increases the pain of patients.Therefore,a fixed limited number of blood collection points(such as 3-5 time points for blood collection)is usually adopted in clinical practice.The area under the MPA concentration curve is calculated by establishing a multiple regression model.However,there are differences in immunization regimens for different patient populations.The accuracy of multiple regression model fitting to calculate mycophenolic acid exposure levels is often not high if fixed and limited blood sampling points are used.To solve the above problems,this paper proposed a prediction method of mycophenolic acid exposure level after kidney transplantation based on few-shot learning.The main research contents include the following three aspects:1)A dynamic selection method of finite blood collection points based on model interpretability technique is proposedIn order to reduce the number of blood collection points for patients after renal transplantation and dynamically select the blood collection points according to the immunization scheme of different patient populations,a dynamic selection method of limited blood collection points based on model interpretability technique is proposed.In this method,SHAP is used to calculate the influence of a single blood collection point as data feature input model on the prediction results and analyze the weight of input features to select the best number of blood collection points for prediction dynamically according to the order of weights.The experimental results show that the blood sampling points selected according to the explicable method can be used to predict MPA-AUC0-12h with high accuracy and good generalization ability,which can effectively help medical workers to analyze the data and get the optimal selection scheme of limited blood sampling points in practical application.It can assist medical staff in clinical diagnosis.2)A method for predicting mycophenolic acid exposure level based on few-shot learning is proposedSince it is difficult to collect clinical data after kidney transplantation,especially after reducing the number of blood collection points after kidney transplantation,the data dimension is further reduced.In order to calculate the correct mycophenolic acid exposure level from the clinical data after low-dimensional renal transplantation,this paper proposes a method for predicting mycophenolic acid exposure level based on few-shot learning based on the dynamic selection of limited blood sampling points that can be interpreted by the model.On the one hand,this method uses Gaussian function which is more suitable for low-dimensional vector aggregation as the attention kernel for the characteristics of low data dimension due to limited blood collection points.On the other hand,in view of the characteristics of less clinical data samples in kidney transplantation,the number of aggregation nodes in the graph attention layer is reduced.The feature extraction effect of the KNN attention pooling layer is improved.The experimental results show that this method has better performance in predicting mycophenolic acid exposure than the traditional methods.When only 10%and 20%samples were used as training samples at 4 blood sampling time points,the mean absolute error of mycophenolic acid exposure level prediction was 12.29%and 13.99%higher than the accuracy of the original model on the test set,respectively.3)A prediction system for mycophenolic acid exposure level after kidney transplantation was designed and implementedA prediction system for mycophenolic acid exposure after kidney transplantation was designed and implemented.Based on WeChat small program development technology,the system provides two functional modules:data collection and mycophenolic acid exposure prediction.Clinicians can input the blood collection data of patients after taking MPA drugs on the system,and calculate mycophenolic acid exposure level by regression method and AI method.