Machine learning applied to molecular identification of Acinetobacter baumannii Global Clone 1

Abstract

ABSTRACTSince the emergence of clones with epidemic behavior, or “high-risk clones”, uninterrupted investigations have been undertaken to understand the genetic and biological basis that led their widespread in nosocomial settings at a global scale. However, its complex genetic and ecological nature only allowed the identification of partial biomarkers with low specificity. The implication of our study is that we successfully applied machine learning algorithms to recognize genetic biomarkers for the identification of Acinetobacter baumannii Global Clone 1, one of the most widespread high-risk clones. Support Vector Machine model predicted U1 sequence with 367 nucleotides length as a specific biomarker for A. baumannii GC1 strains. U1 sequence matched AYE genome in ABAYE1552 locus between 558 and 924 coordinates that corresponds to a fragment of the moaCB gene which encodes a bifunctional protein that includes the molybdenum cofactor biosynthesis protein C and protein B. This region of the gene moaCB allows to differentiate between the A. baumannii GC1 and non-GC1 strains, becoming a suitable biomarker capable of being translated into clinical settings as a molecular typing method such as PCR. Our findings highlight that machine learning can be fruitfully useful even in complex gaps of knowledge of epidemic clones and signify a noteworthy contribution to the literature that could be used to identify challenging nosocomial biomarkers for other multidrug resistant pandemic clones.IMPORTANCEHigh-risk clones play a major role in the spread of resistance. The factors driving the emergence and the dissemination capacity of these clones are not well understood. The dissemination of A. baumannii strains belonging to high-risk GC1 is a global concern. Molecular methods have been used to type GC1 clones such as a particular deletion in intI1 that identify among islands in comM with a Tn6019 backbone. However, since some GC1 strains do not harbor the genomic island AbaR, and intI1 is very common in clinical samples, a more specific biomarker is necessary. The significance of our research is that we successfully applied machine learning methods to predict A. baumannii GC1. Our findings highlight that machine learning can be successfully applied not only to find high-risk clones putative biomarkers but also in disentangling genetic traits that could be used as therapeutic targets to reduce dissemination of lineages with epidemic behavior.