Predicting Human Immunodeficiency Virus (HIV) Drug Resistance Using Recurrent Neural Networks

Abstract

Motivation: Predicting HIV resistance to drugs is one of many problems for which bioinformaticians have implemented and trained machine learning methods, such as neural networks. Predicting HIV resistance would be much easier if we could directly use the three-dimensional (3D) structure of the targeted protein sequences, but unfortunately we rarely have enough structural information available to train a neural network. Furthermore, prediction of the 3D structure of a protein is not straightforward. However, characteristics related to the 3D structure can be used to train a machine learning algorithm as an alternative to take into account the information of the protein folding in the 3D space. Here, starting from this philosophy, we select the amino acid energies as features to predict HIV drug resistance, using a specific topology of a neural network. Results: In this paper, we demonstrate that the amino acid energies are good features to represent the HIV genotype. In addition, it was shown that Bidirectional Recurrent Neural Networks can be used as an efficient classification method for this problem. The prediction performance that was obtained was greater than or at least comparable to results obtained previously. The accuracies vary between 81.3% and 94.7%. Contact:

Drug Resistance Prediction Using Deep Learning Techniques on HIV-1 Sequence Data

“…Drug susceptibility testing results included in the Stanford dataset had been generated using a PhenoSense assay [8]. Susceptibility is expressed as fold change in comparison to wild-type HIV-1; a fold change value greater than 3.5 indicates that a sample is resistant to a given drug [10,17].…”

“…These tests are laborious, time-intensive, and costly. Additionally, phenotypic assays are reliant on prior knowledge of correlations between mutations and resistance to specific drugs, which evolve quickly and thus cannot be totally accounted for [10]. An alternative, the "virtual" or genotypic test, predicts the outcome of a phenotypic test based on the genotype using statistical methods.…”

“…Several machine learning architectures have been applied to predict drug resistance, including random forests such as those used in SHIVA [14][15][16], support vector machines as in geno2pheno [13], decision trees [17], logistic regression [16], and artificial neural networks [10,[18][19][20]. Deep learning models (i.e., neural networks) are a major focus in current machine learning research and have been successfully applied to several classes of computational biology data [21].…”

Probabilistic Neural Network for Predicting Resistance to HIV-Protease Inhibitor Nelfinavir

“…Different techniques have being applied to the development of predictive models, including those based on statistical methods (Prosperi et al, 2009;Van der Borght et al, 2013), neural networks (Bonet et al, 2007;Pasomsub et al, 2010), support vector machine (Beerenwinkel et al, 2003) and decision trees (Beerenwinkel et al, 2002). For the development of such predictive models one has a protein sequence of length n, and since there are 20 amino acids it results in 20 n possible features to represent one sequence.…”

A renaissance of neural networks in drug discovery

“…The theory of RNNs and their application to unsupervised pattern recognition has been described by Orre et al89 This type of neural network has not been used often for drug discovery or modeling of related medical activities or properties.Goh et al first applied RNNs to predicting drug dissolution profiles, and important problem in the pharmaceutical industry 90. More recently, Bonet and coworkers used RNNs to predict HIV drug resistance 91. …”