HIVP-LLM: A Protein-LLM based HIV Drug Susceptibility Predictor

HIVP-LLM is based on protein large language models (e.g., ProteinBERT) to tokeniz HIV protease (HIVP) mutant sequences. It is integrated 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). This program can be practically used to select the most effective treatment strategies for AIDS (acquired immunodeficiency syndrome) patients based on their individual HIVP mutation profiles. This web-based app is implemented with the Flask framework, and has been optimized for friendly use on mobile devices. Concurrently, this pLLM-based approach has also been devised to identify hemolytic peptides and guide further optimization for safe peptidic therapeutics, namely PepHemo-LLM, available here. Also, by combining Chaos Game Representation (CGR) and supervised autoencoder (SAE), we trained an AI model, termed HIVP-CGR, to predict the susceptibility/resistance of AIDS patients with different HIV protease mutations to the 5 drugs, achieving similar robustness and accuracy.

Input: HIVP Mutant Sequence(s)