نوع مقاله : مقاله های پژوهشی
1 دانشجوی کارشناسی ارشد، گروه مهندسی پزشکی، دانشکدهی فنی و مهندسی، دانشگاه اصفهان، اصفهان، ایران
2 استادیار، گروه مهندسی پزشکی، دانشکدهی فنی و مهندسی، دانشگاه اصفهان، اصفهان، ایران
عنوان مقاله [English]
Background: The purpose of this study was to investigate the performance of support vector machines (SVMs) in predicting human immunodeficiency virus (HIV) drug resistance using amino acids sequence analysis.Methods: In order to evaluate the performance of SVMs in predicting HIV drug resistance, SVMs were trained and tested using LibSVM software. Biological methods developed by HIV database (HIVDB), Visible Genetics Inc. (VGI) and the French National Agency for AIDS Research (ANRS) and REGA algorithm were employed to interpret the results of HIV genotypic resistance tests. The results were compared and the most efficient method for each drug was finally determined.Findings: With an accuracy of 86.27- 98.77, SVMs are highly successful classifiers for predicting HIV drug resistance. Using SVMs for HIVDB results has the best performance for amprenavir (APV), nelfinavir (NFV), abacavir (ABC), zidovudine (AZT), stavudine (D4T), didanosine (DDI), tenofovir disoproxil fumarate (TDF), and delavirdine (DLV). Using SVMs for the results of ANRS method has the best performance for indinavir (IDV), lamivudine (3TC), TDF, efavirenz (EFV), and nevirapine (NVP). Using SVMs for REGA results has the best performance for lopinavir (LPV) and AZT. Finally, using SVMs for the results of VGI method has the best performance for IDV, LPV, ritonavir (RTV), saquinavir (SQV), and DDI. Conclusion: SVMs are highly successful classifiers for predicting HIV drug resistance. Before starting treatment with each drug, researchers can determine HIV drug resistance with machine learning methods.