HIVP-CGR: a supervised autoencoder to accurately identify effective drugs for individual patients with HIV protease mutations

HIVP-CGR is based on Chaos Game Representation (CGR) to characterize HIV protease (HIVP) sequences. It is integrated with a supervised autoencoder (SAE) to predict whether a strain of HIV with protease mutations is resistant or susceptible to FDA approved HIVP inhibitors. Here we include 5 drugs: indinavir (IDV), saquinavir (SQV), nelfinavir (NFV), amprenavir (APV) and lopinavir (LPV). The prediction accuracy is so high (95-100%), this program can thus be practically used to guide and select the most effective treatment strategies for AIDS (acquired immunodeficiency syndrome) patients based on individual HIVP mutation profiles. The website is implemented with the Flask framework, and has also been optimized for friendly use on mobile devices. Concurrently, such CGR and SAE combined concept has been further employed to develop PepHemo to model the hemolytic toxicity of peptides. Additionally, a protein large language model HIVP-LLM has also been built to make the predictions, achieving similar high prediction robustness and accuracy.

Input: HIVP Mutant Sequences (FASTA format)


or Upload a file (Example):


Output: Predicted Drug Susceptibility

HIVP Mutant APV IDV LPV NFV SQV


Frequently Asked Questions (FAQ)

HIVP-CGR is a Flask framework-based web server that can be used to predict the susceptibility or resistance of mutant HIV-1 viruses to 5 FDA approved drugs solely based on the mutant protease sequences.

Drug resistance is one of the most pressing issues in HIV treatment today.  The susceptibility testing of HIV-1 drug resistance is essential to developing new anti-viral drugs and optimizing the use of existing drugs. Current methods for HIV resistance testing include phenotypic drug-susceptibility assays to measure drug inhibition of HIV-1 in vitro and genotypic assays that detect mutations known to confer drug resistance. HIVP-CGR is an AI-based method which models the drug susceptibility defined as the ratio of the IC50 of a mutant and a wild-type control. It is implemented as a web server using the Flask framework for quick screening of effective drugs (i.e., protease inhibitors) based on the input of mutant HIV protease sequences.

There are two ways to provide input data to HIV-SAE as described below, and a more detailed instruction is linked to the "Help" button above in the submission form.
  1. In the submission form box on the web page, directly input the HIVP mutant sequences in the fasta format. Examples are provided in the Form.
  2. Upload a file containing all HIVP mutant sequences in the fasta format, with ".fasta" as the file extension. An example file is also provided.
HIVP-CGR currently only accepts 20 natural amino acid types in the fasta format, with each amino acid as a single capital letter. Otherwise, the input will be considered as "illegal". If this happens, an error message will show in the Form. However, the program can handle the sequence input very robustly, for instance, if there are spaces between amino acids, or even blank lines, HIVP-CGR will automatically combine all amino acids together as its input. Please see Help for more details.

HIVP-CGR is free for academic research with a user-friendly interface. Currently, no registration is needed to explore its full functionality. However, commercial users should contact us for agreement and support.

Please cite this web page. The manuscript has been submitted for peer-reviewed publication, and the preprint is currently available on bioRxiv. Therefore, you can also cite: Sequence-based Optimized Chaos Game Representation and Deep Learning for Peptide/Protein Classification. Huang B, Zhang E, Chaudhari R, Gimperlein H. bioRxiv 2022.09.10.507145; doi: https://doi.org/10.1101/2022.09.10.507145

Please contact us at info[at]imdlab[dot]org.