Application of the random forest algorithm to Streptococcus pyogenes response regulator allele variation: from...

Area of Science:

• Microbiology
• Genomics
• Bioinformatics

Background:

• Group A Streptococcus (GAS) is a major global pathogen.
• The emm gene is crucial for GAS genotyping due to its variation and immune selection.
• Understanding GAS genomic traits is vital for epidemiology and control.

Purpose of the Study:

• To apply supervised learning, specifically random forest (RF) algorithms, for inferring GAS genomic traits.
• To evaluate the accuracy of different RF models and response regulator (RR) alleles in predicting traits.
• To explore the plasticity and evolutionary pathways of the GAS mga regulon.

Main Methods:

• Utilized three RF algorithms (Guided, Ordinary, Regularized) with 53 GAS RR allele types.
• Tested inference of six genomic traits: emm-type, emm-subtype, tissue, country, clinical outcomes, and invasiveness.
• Employed feature selection, including minimal sets like mga2 and lrp, to predict emm-type.

Main Results:

• Achieved high accuracy for emm-type prediction (up to 96.7%) using RF and RR alleles.
• Successfully inferred emm-subtype (89.9%), country (88.6%), and invasiveness (84.7%).
• Identified a novel cell wall-spanning domain (SF5) and provided evidence for mga regulon gene excision.

Conclusions:

• RF algorithms effectively infer key GAS genomic traits, advancing genotyping capabilities.
• The study highlights the significant plasticity of the GAS mga regulon and its evolutionary dynamics.
• This workflow enhances the understanding of GAS pathogen biology and evolution.