Computational drug discovery on human immunodeficiency virus with a customized long short-term memory variational...

Area of Science:

• Artificial Intelligence in Drug Discovery
• Computational Virology
• Medicinal Chemistry

Background:

• The human immunodeficiency virus (HIV) remains a global health challenge, with drug resistance and lack of an effective vaccine necessitating novel therapeutic strategies.
• Computational drug discovery offers a powerful approach to identify new anti-HIV agents, complementing traditional methods.

Purpose of the Study:

• To develop and apply a deep-learning model, specifically a long short-term memory (LSTM) variational autoencoder, for the computational discovery of novel HIV therapeutics.
• To generate potential drug candidates and evaluate their efficacy and drug-likeness using established computational models and rules.

Main Methods:

• Training an LSTM variational autoencoder using a dataset of Simplified Molecular Input Line Entry System (SMILES)-encoded compounds.
• Utilizing the trained generative model to design novel molecular structures targeting HIV.
• Assessing the interaction of generated compounds with HIV using a pre-existing artificial intelligence model.
• Evaluating the drug-likeness of novel compounds based on Lipinski's rule of five.

Main Results:

• The LSTM autoencoder achieved a high training accuracy of 91% on a dataset of 1377 compounds.
• The generative model successfully produced novel compounds with predicted anti-HIV activity.
• Generated compounds were found to be drug-like according to Lipinski's rule of five.

Conclusions:

• The study demonstrates the efficacy of a deep-learning LSTM autoencoder for computational drug discovery against HIV.
• This AI-driven approach offers a promising avenue for identifying novel, drug-like candidates in the ongoing fight against HIV.
• The methodology provides a foundation for accelerating the discovery of new anti-HIV therapies.