A Comparison of Different Machine Learning Algorithms for the Prediction of Anti-HIV-1 Peptides Based on Their Sequence-Related Properties

作者：Poorinmohammad Naghmeh; Mohabatkar Hassan^*

来源：International Journal of Peptide Research and Therapeutics, 2015, 21(1): 57-62.

DOI：10.1007/s10989-014-9432-x

• Human immunodeficiency virus-1
• Machine learning
• Anti-HIV-1 peptide
• Multilayer perceptron
• Logistic model tree

Human immunodeficiency virus-1 (HIV-1) is the etiological agent for the global concerning disease "AIDS". The virus infects 35 million people globally and after 30 years, the disease remains a challenge. Despite great efforts in finding efficient treatment strategies, the pandemic of AIDS is continuing and the rate of new infections has not been diminished. Therefore, the need for finding novel treatment strategies is still of great importance. Peptide-based therapeutics has shown promise in the treatment of many challenging diseases such as various types of cancers and also HIV. Since time and money are the two restricting factors in any experimental researches, computer-aided techniques can dramatically reduce time and costs. In the present study, we developed a method based on pseudo amino acid composition of amino acid sequences to classify the anti-HIV-1 peptides using different machine learning algorithms. The performance of each algorithm was investigated and after comparing the performance parameters, the most accurate algorithm was proposed for predicting anti-HIV-1 activity of any given peptide. Having the accuracies of 96.15 and 83.71 % respectively, multilayer perceptron (MLP) and logistic model tree algorithms were primarily shown to be the most accurate ones in classifying anti-HIV-1 peptides. Final results demonstrate that model generated by MLP can be a valuable tool for the classification and prediction of anti-HIV-1 peptides in order to have a preliminary prediction which can be further coupled with experimental assays while reducing time and costs.

• 出版日期2015-3