Deep-learning-enabled antibiotic discovery through molecular de-extinction.

Area of Science:

• Biochemistry
• Genomics
• Computational Biology

Background:

• Antibiotic resistance is a growing global health threat.
• Traditional methods for discovering new antibiotics are slow and often unsuccessful.
• Extinct organisms may possess unique molecules with therapeutic potential.

Purpose of the Study:

• To explore molecular de-extinction for discovering novel antibiotic peptides.
• To leverage deep learning for mining proteomes of extinct organisms.
• To identify and validate new antimicrobial compounds.

Main Methods:

• Trained deep learning models to predict antimicrobial activity of peptide sequences.
• Mined proteomes of extinct organisms, analyzing over 10 million peptides.
• Synthesized and experimentally tested predicted antibiotic peptides.
• Investigated the mechanism of action and in vivo efficacy of lead compounds.

Main Results:

• Identified 37,176 predicted antimicrobial sequences, with 11,035 novel sequences.
• Experimentally confirmed activity of 69 synthesized peptides against bacterial pathogens.
• Discovered that most novel peptides kill bacteria by depolarizing the cytoplasmic membrane.
• Demonstrated in vivo anti-infective activity of lead compounds in mouse models.

Conclusions:

• Deep learning-assisted molecular de-extinction is a viable strategy for discovering novel therapeutics.
• Ancient peptides offer a new class of antimicrobials with unique mechanisms of action.
• This approach holds significant potential for accelerating the development of treatments for antibiotic resistance and other diseases.