Diabetes mellitus or diabetes is a kind of disease characterized by the raised of blood sugar. This disease can deal long-term damage, such as dysfunction, and failure of various organs. In Indonesia, diabetes is one of a major causes of death with more than 10 million people living with diabetes. To date, there is no drug that can cure diabetes. So far, people with diabetes must take responsibility for their daily routine. Drug discovery is needed to find the cure of diabetes. PTP1B is one of inhibitor that has been proved as a promising target for anti-diabetes mellitus. Drug discovery takes a lot of time and effort, thus, in the silico methods, such as quantitative structure-activity relationship (QSAR), can be used to accelerate this process. We aim to build a QSAR model of PTP1B inhibitor as anti-diabetes melitus using simulated annealing (SA)-Support Vector Machine (SVM) method. The data were retrieved from ChEMBL database by selecting the SMILES from each compound, by calculating the SMILES using PaDEL, we got 1443 descriptors for each compound, and using SA, we will decrease the number of descriptors. The best result shows that SA selected 600 descriptors out of 1443 descriptors for each compound. The use of RBF kernel on SVM has the best value with accuracy, F1 score, and AUC of 94.508%, 95.048%, and 0.943, respectively.
Keywords: quantitative structure-activity relationship (QSAR), diabetes mellitus, protein tyrosine phosphatase 1B (PTP1B), simulated annealing (SA), support vector machine (SVM)