Machine Learning Accelerates De Novo Design of Antimicrobial Peptides

Area of Science:

• Computational chemistry
• Biotechnology
• Machine learning in drug discovery

Background:

• Antimicrobial peptides (AMPs) are crucial for developing new treatments against resistant pathogens.
• De novo design of AMPs using computational methods offers a promising avenue for drug discovery.

Purpose of the Study:

• To develop and validate a machine learning-based model (AP_Sin) for accurate prediction of AMPs.
• To create a peptide sequence generator (AP_Gen) for de novo AMP design.
• To accelerate the discovery of novel AMPs with potent antimicrobial activities.

Main Methods:

• Trained AP_Sin using 1160 AMP and 1160 non-AMP sequences.
• Employed AP_Gen for de novo generation of 17,496 tridecapeptide sequences.
• Screened generated sequences with AP_Sin, identifying 2675 candidates.
• Chemically synthesized 180 candidate AMPs for experimental validation.

Main Results:

• AP_Sin achieved 94.61% accuracy in AMP classification, outperforming existing models.
• 18 out of 180 synthesized peptides demonstrated significant antimicrobial activity.
• 16 peptides exhibited minimal inhibitory concentrations (MIC) below 10 μg/mL against tested pathogens.

Conclusions:

• The developed computational approach, combining AP_Sin and AP_Gen, effectively accelerates the identification of novel antimicrobial peptides.
• This research provides a novel and efficient strategy for the de novo design of AMPs with therapeutic potential.