Variational Autoencoder-enabled High-throughput Drug Screening for HIV Latency Modulators predicted through Noise in...

Area of Science:

• Computational biology
• Virology
• Drug discovery

Background:

• Human immunodeficiency virus (HIV) latency, characterized by reservoirs of dormant infected cells, presents a major obstacle to achieving a cure.
• Current strategies like "shock and kill" and "block and lock" aim to control viral reactivation, but require effective latency modulators.
• Previous drug screens linked gene expression noise modulation to HIV latency modulation, but these methods are resource-intensive.

Purpose of the Study:

• To develop a cost-effective and efficient method for identifying compounds that modulate human immunodeficiency virus (HIV) latency.
• To leverage existing large-scale experimental data for in silico drug screening.
• To validate the efficacy of computationally predicted HIV latency modulators.

Main Methods:

• Trained a variational autoencoder (VAE) on a large-scale time-lapse fluorescence microscopy dataset of HIV gene expression.
• Performed in silico screening of approximately 175,000 compounds to predict potential HIV latency modulators.
• Experimentally tested the top 113 predicted compounds for their ability to modulate HIV latency.

Main Results:

• Identified 16 latency reversing agent (LRA) synergizers and 2 latency promoting agents (LPAs) among the tested compounds.
• Achieved an overall experimental hit rate of 15.9% for predicted HIV latency modulators.
• Demonstrated the effectiveness of in silico screening guided by experimental data in identifying potent modulators.

Conclusions:

• In silico drug screening using VAEs trained on existing datasets offers a high-throughput and cost-effective approach to discover HIV latency modulators.
• This computational strategy significantly reduces the reliance on labor-intensive wet lab methodologies.
• The validated compounds show promise for advancing the development of novel therapeutic strategies for HIV cure.