Pangenomic and machine learning approaches to divergent pathogenic strategies in avian Mycoplasma species: evidence and limitations.

Abstract

Avian mycoplasmosis, mainly caused by Mycoplasma gallisepticum (MG), Mycoplasma synoviae (MS), and Mycoplasma gallinarum (MGn), remains a persistent constraint on poultry production. These species co-circulate in flocks but differ in tissue tropism, host responses, and clinical outcomes. This review examines how pangenomic analysis and machine learning (ML)-based feature prioritization can clarify these differences while separating computational associations from experimentally supported mechanisms. We first summarize genome reduction, structural variation, accessory gene content, and antimicrobial resistance architecture, and then evaluate ML applications in classification, feature selection, network inference, and candidate prioritization. The evidence is strongest for MG, which combines respiratory adhesion, TLR2/MyD88/NF-κB-mediated inflammation, NLRP3 activation, and oxidative injury. MS appears more synovium-adapted, with evidence of fibroblast reprogramming, ER stress-related responses, and interference with antiviral signaling. MGn remains insufficiently characterized because complete reference genomes, standardized infection models, and functional validation data are lacking. We conclude that pangenomics and machine learning are useful only when linked to validation: They can define candidate repertoires and rank testable hypotheses, but they do not establish pathogenic mechanisms. A controlled three-species comparison and a complete, well-annotated MGn reference genome are the most immediate priorities for addressing the gaps that currently prevent confident cross-species comparison.