A machine learning method for predicting molecular antimicrobial activity.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Machine Learning

Background:

• Antibiotic resistance is a growing global health threat.
• Traditional antibiotic discovery methods face limitations.
• Novel computational approaches are needed to accelerate drug discovery.

Purpose of the Study:

• To introduce MFAGCN, a machine learning model for predicting antimicrobial efficacy.
• To leverage molecular fingerprints (MACCS, PubChem, ECFP) and graph representations for enhanced prediction.
• To identify influential molecular functional groups in antimicrobial activity.

Main Methods:

• Developed MFAGCN, a graph convolutional network model with an attention mechanism.
• Integrated multiple molecular fingerprints (MACCS, PubChem, ECFP) and molecular graph features.
• Conducted comparative experiments against baseline models on public datasets.
• Performed functional group distribution analysis and structural similarity analysis with known antibiotics.

Main Results:

• MFAGCN demonstrated superior performance compared to baseline models on two public datasets.
• Analysis validated the model's predictions and highlighted the importance of functional groups.
• Structural similarity analysis successfully prevented the rediscovery of existing antibiotics.

Conclusions:

• MFAGCN offers a rapid and effective method for screening molecules with antimicrobial potential.
• The model provides valuable insights into functional groups driving antimicrobial activity.
• This approach accelerates the development of novel antibiotics to combat resistance.