ProbML: A Machine Learning-Based Genome Classifier for Identifying Probiotic Organisms

Abstract

Probiotics are microorganisms that offer health benefits to the host. Traditional methods for identifying these organisms are time-consuming and resource-intensive. This study addresses the need for a more efficient and accurate approach to probiotic identification using machine learning (ML) techniques. The present study introduces ProbML, an ML-based approach for identifying probiotic organisms from whole genome sequences of prokaryotes. Among the five ML algorithms tested, XGBoost models demonstrated superior performance, achieving a maximum accuracy of 100% on learning data and 95.45% on an independent test dataset. This surpasses existing tools, which achieved 97.77% and 66.28% accuracy on the same datasets, respectively. The ProbML models were used to analyze 4728 genomes in the Unified Human Gastrointestinal Genome database, classifying 650 genomes as probiotics, with many previously unreported. A versatile GUI platform was also developed that employs ProbML models for probiotic classification or can be used to generate custom ML classifiers based on user-specific needs (https://github.com/sysbio-iitmandi/MLG_Dashboard). This study emphasizes the power of genomic data and advanced ML techniques in accelerating probiotic discovery.