Convolutional Networks for predicting Antimicrobial Resistance

Antibiotic resistance is a growing public health concern worldwide. Many diagnostic techniques were developed, but none of them are fast enough to predict the presence of bacteria with a resistant phenotype. This work proposes the use of a convolutional neural network on bacterial DNA sequences to predict resistance. Once the optimal neural network architecture was selected, validation was performed using a k-fold procedure estimating the loss function of both validation and test and evaluating the confusion matrix. Thanks to this technology it would be possible to speed up the choice of a proper therapeutic strategy and to avoid the rise of untreatable infection diseases.