Proof of concept: artificial intelligence clustering of Burkholderia species according to antibiograms

Abstract

There are challenges to designing an automated system for screening and categorising infectious bacterial species. One of the challenges is clustering and visualisation of data. Principle Component Analysis (PCA) is a widely used algorithm used in machine learning, which maximizes divergences among clusters of data on common planes. Antibiograms are available as tables, which can be converted to mathematical matrices. PCA can use these converted matrices and find clusters of antibiograms without supervision. To show the utility of PCA in clustering antibiograms, Burkholderia species are chosen for the following reasons. First, Burkholderia species have variety of in vitro antibiograms among species and within a single species. Burkholderia cepacia complex is a heterogenous group of the Burkholderia causing respiratory diseases in paediatric patients with cystic fibrosis and understanding in vitro antibiograms of the species can be useful to optimising antibiotic regiment. Here, PCA is used to demonstrate antibiogram clusters of Burkholderia multivorans, a member of Burkholderia cepacia complex. Resources: GNU Octave (https://octave.org) PCA (https://nbviewer.org/github/susilvaalmeida/machine-learning-andrew-ng/blob/master/Programming%20Exercise%207%20-%20K-means%20Clustering%20and%20Principal%20Component%20Analysis.ipynb)