Predição de resistência antimicrobiana em pseudômonas aeruginosa com aprendizagem de máquina

Abstract

The drop in the production of new antibiotics along with the increase in resistant bacteria, proved necessary to find alternative ways to combat this problem. The recommendation given by the World Health Organization is the awareness in the administration of antibiotics to patients in need. In this aspect, machine learning has been shown in recent years as an excellent solution. With the objective of combating the multiresistant bacteria Pseudomonas aeruginosa, samples were obtained through the well-known public database PATRIC. Treatment of data through k-mer analysis proved to be effective, but computationally costly, limiting the scope of the research. Moving foward, machine learning algorithms training was carried out with Adaboost, Bagging, SVM, Random Trees, Logistic Regression and GradientBoost. The e F1-score averaging between 0.7-0.8, highlights the potential of these tools to help the treatment of bacterial infections.