Machine learning for predicting antimicrobial efficacy of periodontal gel formulations in vitro biofilm models

TL;DR

This paper uses machine learning to predict how well periodontal gels work against biofilms, speeding up the development of effective treatments.

Contribution

The novel use of Gradient Boosting machine learning to predict antimicrobial efficacy of periodontal gels with high accuracy.

Findings

Gradient Boosting achieved 92.8% accuracy in predicting antimicrobial efficacy of periodontal gels.

Key predictors included antimicrobial type, concentration, and polymer viscosity.

ML models can reduce the need for extensive in vitro testing of gel formulations.

Abstract

Periodontal disease caused by dysbiotic biofilms poses a major challenge and predicting the efficacy of topical antimicrobial gels is
limited by biofilm resistance and resource-intensive in vitro testing. Therefore, it is of interest to develop machine
learning (ML) models to predict antimicrobial efficacy of novel gel formulations against multi-species periodontal
biofilms. Hence, a total of 120 formulations with varying polymers, agents, concentrations and enhancers were tested using the Calgary
Biofilm Device and efficacy data were used to train Random Forest, SVM, Gradient Boosting and Neural Network models. Gradient Boosting
achieved the best performance (accuracy 92.8%, AUC-ROC 0.96), with antimicrobial type, concentration and polymer viscosity
as key predictors. ML, particularly Gradient Boosting, offers a reliable tool for predicting periodontal gel efficacy, enabling faster
formulation optimization and reducing the need for extensive laboratory screening.