A Deep Learning Approach to Antibiotic Discovery.

Area of Science:

• Microbiology
• Artificial Intelligence
• Drug Discovery

Background:

• The rise of antibiotic-resistant bacteria necessitates the discovery of novel antimicrobial agents.
• Conventional antibiotic discovery methods face significant challenges in identifying new drug candidates.

Purpose of the Study:

• To develop and apply a deep neural network for predicting molecules with antibacterial activity.
• To identify novel, structurally distinct antibacterial compounds using artificial intelligence.

Main Methods:

• Training a deep neural network on chemical libraries to predict antibacterial properties.
• Screening large chemical databases, including the ZINC15 database, for potential antibiotic candidates.
• Testing identified compounds, such as halicin, for efficacy against a broad spectrum of bacterial pathogens in vitro and in vivo.

Main Results:

• Discovery of halicin, a molecule structurally different from existing antibiotics, exhibiting broad-spectrum bactericidal activity.
• Successful treatment of Clostridioides difficile and pan-resistant Acinetobacter baumannii infections in murine models using halicin.
• Identification of eight novel antibacterial compounds from over 107 million molecules, all structurally distinct from known antibiotics.

Conclusions:

• Deep learning models are effective tools for accelerating antibiotic discovery.
• AI can identify novel antibacterial molecules with unique structures, expanding therapeutic options against resistant pathogens.
• This approach offers a promising strategy to replenish the dwindling antibiotic pipeline.