AI Discovers a New Antibiotic That Kills Superbugs

 

Researchers at Massachusetts Institute of Technology have discovered a new antibiotic using artificial intelligence. The antibiotic, named halicin, shows strong activity against multiple drug-resistant bacterial pathogens and represents a new approach to antibiotic discovery.

The study demonstrates how machine learning can accelerate drug discovery and help address global antibiotic resistance.

The research team developed a deep learning model trained to identify chemical compounds capable of inhibiting bacterial growth. The system analyzed thousands of molecules and predicted antibacterial activity based on chemical structure.

Unlike traditional drug discovery methods, which require extensive laboratory screening, the AI model rapidly evaluated large chemical libraries and identified candidate molecules with potential antimicrobial effects.

One compound identified by the model showed strong antibacterial properties and was later named halicin.

Mechanism of Action of Halicin

Halicin functions differently from conventional antibiotics.

Most antibiotics target:

• bacterial cell wall synthesis
• protein production
• DNA replication

Halicin instead disrupts the electrochemical gradient across bacterial cell membranes. This prevents bacteria from maintaining energy production processes required for survival, leading to cell death.

Because of this novel mechanism, bacteria show limited ability to develop resistance against the drug.

Activity Against Resistant Pathogens

Laboratory testing showed that halicin is effective against several clinically important pathogens, including:

• Escherichia coli
• Mycobacterium tuberculosis
• Carbapenem-resistant Enterobacteriaceae
• Acinetobacter baumannii

In mouse infection models, halicin successfully cleared infections caused by drug-resistant A. baumannii within 24 hours.

Researchers also observed minimal resistance development during repeated exposure experiments. The study demonstrates that artificial intelligence can identify structurally unique antibiotics and significantly accelerate drug discovery. The findings suggest that machine learning tools may help overcome limitations of traditional antibiotic development and support future antimicrobial research.

References / Sources

A Deep Learning Approch To Antibiotics - MIT