Application of the random forest algorithm to Streptococcus pyogenes response regulator allele variation: from machine learning to evolutionary models

Abstract

Group A Streptococcus (GAS) is a globally significant bacterial pathogen. The GAS genotyping gold standard characterises the nucleotide variation of emm, which encodes a surface-exposed protein that is recombinogenic and under immune-based selection pressure. Within a supervised learning methodology, we tested three random forest (RF) algorithms (Guided, Ordinary, and Regularized) and 53 GAS response regulator (RR) allele types to infer six genomic traits (emm-type, emm-subtype, tissue and country of sample, clinical outcomes, and isolate invasiveness). The Guided, Ordinary, and Regularized RF classifiers inferred the emm-type with accuracies of 96.7%, 95.7%, and 95.2%, using ten, three, and four RR alleles in the feature set, respectively. Notably, we inferred the emm-type with 93.7% accuracy using only mga2 and lrp. We demonstrated a utility for inferring emm-subtype (89.9%), country (88.6%), invasiveness (84.7%), but not clinical (56.9%), or tissue (56.4%), which is consistent with the complexity of GAS pathophysiology. We identified a novel cell wall-spanning domain (SF5), and proposed evolutionary pathways depicting the ‘contrariwise’ and ‘likewise’ chimeric deletion-fusion of emm and enn. We identified an intermediate strain, which provides evidence of the time-dependent excision of mga regulon genes. Overall, our workflow advances the understanding of the GAS mga regulon and its plasticity.

Lessons Learnt From Using the Machine Learning Random Forest Algorithm to Predict Virulence in Streptococcus pyogenes

“…We observed a strain-dependent variability in the IGRs and coding sequences of GAS TCSs and TRs using phylogenetics and concordance, and proposed a set of core TRs as candidates for a novel GAS typing system. These subsequently informed the design of our ML workflow, in which we were able to predict the GAS strain (emm type) with 97% accuracy and establish that mga2 and lrp were the most mathematically powerful predictors of strain (Buckley et al, 2021). Overall this finding was important because it revealed a backward-compatibility between our TRbased typing system and the vast emm-based knowledge set.…”

“…Collectively, these findings were significant because it has been suggested (Lees et al, 2019) that an ability to detect rare genotype anomalies enhances the discovery of rare clinically-relevant phenotypes. We were also able to predict the country of origin using this approach, suggesting a geography-dependent evolution of GAS TRs (Buckley et al, 2021).…”

“…We discovered examples of rare recombination of the mga regulon including mga2-switching, emm-switching, and chimeric emm-enn events and were also able to develop evolutionary models to explain them (Buckley et al, 2020;Buckley et al, 2021). Furthermore, we identified the deletion of a transporter gene (maeP) that stands as a biomarker for the invasive emm subtype 89.0 (Buckley et al, 2018).…”

“…We were able to predict strain and geography with high accuracy, but were unable to predict virulence phenotype. By investigating the causes of inaccuracy in examples of where the predicted strain differed from the published strain, we were able to discover several rare anomalies in the mga regulon (Buckley et al, 2021). We identified a novel cell-wall spanning domain (SF5) which is described as a chimera of SF3 and SF1, that redefines the GAS emm pattern typing system.…”

“…We recently characterised the distribution and diversity of the nucleotide sequences of the two-component systems (TCSs) and stand-alone transcription regulators (TRs) in 944 GAS genomes, and then explored phenotype associations using phylogeny and concordance metrics (Buckley et al, 2018;Buckley et al, 2020). Subsequently, we applied the ML random forest (RF) algorithm to the allelic variation of the TRs to predict six metadata traits of the genomes (Buckley et al, 2021). These were emm type, emm subtype, country and tissue of the sample, propensity to cause invasive disease, and clinical outcome.…”