Explainable deep-learning techniques for the study of antibiotic resistance in bacterial infectants

AMR has been listed as one of the leading public health threats of the 21st century, with some estimates predicting it to cause the deaths of 10 million people a year by 2050. WHO and numerous other groups and researchers agree that the spread of AMR is an urgent issue requiring a global, coordinated action plan to address, and more and more publications are addressing the issue and estimating the ARM burden in terms of health and public health expenses. State-of-the-art sequencing techniques (next generation sequencing, NGS) have made it possible in recent years to obtain the genome sequence of an organism in short time and at low cost, thus enabling a study of resistance through DNA analysis. In particular, promising results have been obtained through Machine Learning techniques applied to the gene sequence to identify the parts responsible for drug resistance. The technique required to investigate the reasons for a decision made by a deep learning model is called Explainable AI (XAI). XAI is very important for critical applications, such as such as defence, health care, law and order, and autonomous driving vehicles, where it is required for safety and transparency reasons. In the medical field, it could help the research towards a better understanding of the mutations that can cause the resistance. The purpose of this thesis is to create a system that traces, given a prediction of resistance to an antibiotic from DNA sequences, to the portions of the DNA that are the network's identified cause for resistance.