Predicting Antimicrobial Activity of Conjugated Oligoelectrolyte Molecules via Machine Learning

Area of Science:

• Drug Discovery and Development
• Computational Chemistry
• Antimicrobial Resistance

Background:

• Growing antibiotic resistance necessitates novel drug development, a process traditionally taking decades.
• Existing machine learning models for drug discovery often require large datasets and focus on known antibiotic structures.
• Unconventional antibiotic classes present challenges for traditional structure-activity relationship studies.

Purpose of the Study:

• To develop a predictive model for antimicrobial activity of conjugated oligoelectrolyte molecules, a novel antibiotic class.
• To identify key molecular descriptors for predicting antimicrobial efficacy.
• To demonstrate a machine learning approach adaptable to other new antibiotic domains.

Main Methods:

• Developed a machine learning model to predict minimum inhibitory concentration (MIC) against *E. coli* K12.
• Utilized recursive elimination to select 21 relevant molecular descriptors from an initial set of 5305.
• Employed a molecular representation reflecting the three-dimensional shape of conjugated oligoelectrolytes.

Main Results:

• Achieved a predictive model with an R-squared value of 0.65 for antimicrobial activity.
• Identified optimal molecular representation, emphasizing 3D shape, as critical for prediction accuracy.
• Demonstrated successful prediction without prior knowledge of the specific antimicrobial mechanism.

Conclusions:

• A robust machine learning model can predict antimicrobial activity for novel antibiotic classes like conjugated oligoelectrolytes.
• The choice of molecular representation is paramount for accurate antimicrobial activity prediction.
• This approach offers a scalable method for accelerating the discovery of new antibiotics against resistant pathogens.