Using Machine Learning to Predict the Antibacterial Activity of Ruthenium Complexes

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Antimicrobial Drug Discovery

Background:

• Rising antimicrobial resistance (AMR) necessitates novel antibiotic development pipelines.
• Metal complexes show promise as antimicrobials, but research is limited compared to organic molecules.
• Machine learning (ML) is increasingly used for small organic molecule design, even with limited data.

Purpose of the Study:

• To apply ML for the first time in discovering metal-based antibacterial agents.
• To train ML models using ruthenium arene Schiff-base complexes and their antibacterial data.
• To predict and validate the activity of novel metal-based compounds.

Main Methods:

• Synthesized 288 ruthenium arene Schiff-base complexes.
• Assessed antibacterial properties of synthesized complexes.
• Trained ML models on the dataset of complexes and their activity.
• Predicted activity of 54 new compounds using the trained ML models.

Main Results:

• ML models demonstrated strong predictive performance.
• Predicted compounds showed a 5.7x higher hit rate (53.7%) against methicillin-resistant Staphylococcus aureus (MRSA) compared to the initial library (9.4%).
• Successful application of ML in identifying potent metalloantibiotics.

Conclusions:

• ML can significantly enhance success rates in the search for new metalloantibiotics.
• This study validates ML as a powerful tool for metal-based drug discovery.
• Paves the way for broader ML applications in designing metal-based medicines.