A Model for Hepatotoxicity Prediction Based on Coarse-Grained Parallel Genetic Algorithm and Support Vector Machine

作者：Ding, Sha; Zhao, Shi-Yuan; Chen, Zhi; Lin, Tao^*

来源：Journal of Medical Imaging and Health Informatics, 2016, 6(8): 1896-1903.

DOI：10.1166/jmihi.2016.1976

• Prediction of Hepatotoxicity
• Support Vector Machine
• Parallel Genetic Algorithm
• Parameter Optimization
• Feature Selection

One of the prime reasons for the patients suffering from the acute liver damage or dysfunction is drug. Thus, it is imperative to carry out the further research on hepatotoxicity affected by the drugs before the drugs enter the market. In order to obtain better classification accuracy of hepatotoxicity prediction, support vector machine (SVM) associates with a coarse-grained parallel genetic algorithm (CGPGA) to complete the process of selecting feature and optimizing parameter of SVM simultaneously. The proposed model shortly named as CGPGA-SVM approach. To compare with the previous learning algorithms, the same dataset is necessary which includes 1087 compounds. Firstly, in the total dataset with the external validation, the overall accuracy for the prediction is 75.63% by 10-fold cross-validation. As the result, the figure is 12% higher than the previous congeneric prediction. Next, training set and test set are produced to form the resource in this experiment. The final prediction accuracies are 78.21% and 70.89% by CGPGA-SVM approach. From the aspect of the efficiency, there is a contrast among the serial genetic algorithm (SGA), Grid search and CGPGA-SVM by the program. Meanwhile, Our proposed approach reduces the running time obviously. The result implicates the performance of the proposed CGPGA-SVM approach is better in hepatotoxicity prediction.

• 出版日期2016-12
• 单位四川大学