The spread of antimicrobial resistant organisms, and the extension of specific species being resistant to a broader range of antimicrobials, continues to present a considerable threat to the hospital setting. Many species are nosocomial infectious agents, increasingly difficult to treat, and posing a particular threat to immunocompromised patients. Multiple different genes confer resistance to a given antimicrobial agent.
Understanding regional variations in antimicrobial resistance has a two-fold importance. First, it enables scientists to understand the spread of resistance and to alert about the loss of efficacy of a particular agent to a given bacterial species .Second, it aids medical professionals in deciding which antimicrobial to administer to patient. Often there is a little time to characterise the infectious species in order to determine the optimal antimicrobial. By understanding patterns of resistance in the community, some antimicrobials may be preferential to others at the local level.
One means to advance regional-centric understanding of antibiotic resistant patterns is through the use of machine learning to make computational predictions. This form of artificial intelligence provide an algorithm, with the ability to predict certain outcomes through a learned model by providing a large amount of experimental data. A portion of these data are training data, used to increase the success rate of the predictions made. Once the cross-validation score (‘training set’) has reached an acceptable level, real world clinical data can be scrutinized (‘testing set’).
Such analyses can reveal hitherto concealed antimicrobial resistance determinants by scrutinizing metagenomics datasets, datasets of environmental microbiomes and their pathogenic potential in humans. In addition, the twinning of machine learning algorithms and laboratory testing can aid in the acceleration of discovering new antimicrobials. This latter step involves computer-aided prospection to align novel drugs with alternative mechanisms of antimicrobial action (what are called ‘synergistic medication combinations’).
Machine learning approaches include logistic regression (LR), support vector machine (SVM), random forest (RF) and convolutional neural network (CNN).
Machine learning algorithms are able to correlate genomic variations with phenotypes and look for patterns of resistance against given agents within regions. Scrutinizing such databases involves an algorithm using conjunction (logical-AND) or disjunction (logical-OR) Boolean functions.
Researchers based at the Jeffrey Cheah Biomedical Centre, Wellcome-MRC Cambridge Stem Cell Institute, at the University of Cambridge, U.K., have been developing such a model and the results have been published in the journal Microbiome. The research undertakes predictive analysis at the very local, niche level of the International Space Station (“Machine learning algorithm to characterize antimicrobial resistance associated with the International Space Station surface microbiome”).
