(HIV drug Resistance) web service is a freely available tool on the Way2Drug platform that predicts HIV-1 resistance to antiretroviral drugs from amino acid sequences of viral protease or reverse transcriptase obtained from clinical samples. It was developed on the basis of machine learning models trained with genotype–phenotype data from the Stanford HIV Drug Resistance Database.
The service accepts amino acid sequences of HIV-1 reverse transcriptase or protease directly as user input. Prior to prediction, each submitted sequence undergoes automatic alignment to a reference wild-type sequence using a modified Smith–Waterman algorithm, ensuring robust handling of sequence variants, insertions, deletions, and ambiguous positions from clinical sequencing data. The service then applies the best-performing models for individual drugs, using two complementary approaches: k-mer–based models with Random Forest or Support Vector Machine algorithms, and MultiPASSR models based on multi-n-gram descriptors and a Naive Bayes method.
HVR employs two complementary machine learning approaches to generate resistance predictions:
K-mer models - amino acid subsequences of optimized length (5–32 residues) are extracted from the full enzyme sequence and encoded as binary feature vectors; models are built using Random Forest and Support Vector Machine algorithms from the scikit-learn library;
Multi-n-gram models (MultiPASSR) - string descriptors of length 1–15 are generated from the sequence and evaluated using a Naive Bayes approach implemented in MultiPASS software;
An additional consensus prediction option combines the outputs of both approaches using the logical operator "AND," further improving reliability.
The service provides resistance predictions for 18 antiretroviral drugs across three therapeutic classes:
Protease Inhibitors (PI) - Fosamprenavir, Atazanavir, Indinavir, Lopinavir, Nelfinavir, Saquinavir, Tipranavir, Darunavir;
Nucleoside RT Inhibitors (NRTI) - Lamivudine, Abacavir, Zidovudine, Stavudine, Didanosine, Tenofovir;
Non-nucleoside RT Inhibitors (NNRTI) - Efavirenz, Nevirapine, Etravirine, Rilpivirine.
All classification models were validated using 5-fold cross-validation on genotype–phenotype datasets sourced from the Stanford HIV Drug Resistance Database. The models demonstrate high reliability. Balanced accuracy ranges from 0.82 to 0.94 across all drugs. For protease inhibitors such as nelfinavir, the area under the curve (AUC) values reach up to 0.97. Performance is particularly strong for protease inhibitor predictions, where balanced accuracy consistently exceeds 0.84.
HVR can be used as a supportive tool for the interpretation of HIV genotyping results in research and clinical practice. By analyzing amino acid sequences of HIV protease or reverse transcriptase, the service helps estimate possible resistance to specific antiretroviral drugs and may assist in selecting more effective treatment options.
The service is especially useful when rapid preliminary assessment of resistance is needed and phenotypic testing is limited or unavailable. It can also support comparative analysis of viral variants, educational tasks, and studies focused on mutation patterns associated with HIV drug resistance.
Rapid resistance screening - the service delivers resistance predictions for 18 antiretroviral drugs in seconds, making it suitable for quick preliminary assessment of viral samples from clinical or research sequencing workflows.
Broad drug class coverage - predictions span all three major antiretroviral classes (protease inhibitors, NRTIs, and NNRTIs), providing a comprehensive resistance profile from a single sequence submission.
Improved prediction confidence - by combining two independent machine learning approaches (k-mer and multi-n-gram models) through a consensus prediction mode, the service reduces the risk of false-positive resistance calls.
Anastasiia I. Paremskaia et al. (2019)
Web Service for HIV Drug Resistance Prediction Based on Analysis of Amino Acid Substitutions in Main Drug Targets.
Viruses, 15(11):2245.
doi: 10.3390/v15112245
If you need to evaluate a large dataset, or if you need to maintain confidentiality of structural formulas transmitted via unsecured data channels, you can contact us to discuss the licensing opportunities.
© 2011 - 2026 Way2Drug. All rights reserved.