De Novo–HIV: Neural-Quantum Protease Inhibition

Abstract

The search for potent drug candidates targeting the Human Immunodeficiency Virus (HIV) continues to be a significant global health challenge, largely because of the virus’s rapid mutation rate, the emergence of drug resistance, and the lengthy, expensive nature of traditional drug development processes. In recent years, artificial intelligence (AI) and deep learning have shown great promise in expediting early-stage drug discovery. Notably, Long Short-Term Memory (LSTM) networks have demonstrated remarkable ability in understanding chemical representations and generating new molecular structures through sequence-based notations like SMILES. This paper presents a structured review of LSTM-based and related deep learning approaches applied to HIV drug discovery. Existing studies are critically analysed and classified based on their learning objectives, molecular representation strategies, and validation mechanisms. Key research gaps are identified, including limited generative diversity, lack of multi-objective optimization, and insufficient biological validation. Finally, a conceptual hybrid framework is discussed that integrates LSTM-based molecular generation with advanced evaluation strategies, offering future research directions for scalable and clinically relevant HIV drug discovery.