Predicting drug resistance in M. tuberculosis using a Long-term Recurrent Convolutional Networks architecture
AbstractDrug resistance in Mycobacterium tuberculosis (MTB) may soon be a leading worldwide cause of death. One way to mitigate the risk of drug resistance is through methods that predict drug resistance in MTB using whole-genome sequencing (WGS) data. Existing machine learning methods for this task featurize the WGS data from a given bacterial isolate by defining one input feature per SNP. Here, we introduce a gene-centric method for predicting drug resistance in TB. We define one feature per gene according to the number of mutations in that gene in a give isolate. This representation greatly decreases the number of model parameters. We further propose a model that considers both gene order through a Long-term Recurrent Convolutional Network (LRCN) architecture, which combines convolutional and recurrent layers. We find that using these strategies yields a substantial, statistically-significant improvement over the state-of-the-art, and that this improvement is driven by the order of genes in the genome and their organization into operons.