The use of variational autoencoders to characterise the heterogeneous subpopulations that arise due to antibiotic treatment

Abstract

Antimicrobial resistance (AMR) is a persistent threat to global agriculture and healthcare systems. One of the challenges towards development of robust antimicrobials to date has been the limitation posed by low resolution bacterial sequencing technologies. The recent development of Bacterial Single Cell RNA sequencing protocols has provided an unprecedented opportunity in AMR research as it now enables researchers to probe bacterial populations at single cell resolution. In this study, we apply a Bayesian Variational Autoencoder, MrVI, to data generated by one such Bacterial Single Cell RNA sequencing protocol, BacDrop, and use it characterise changes in gene expression levels before and after antibiotic perturbation. Through the use of MrVI, we were able to find distinct DNA damage and heat shock response subpopulations. We also determined that each of the subpopulations could be mapped back to its respective antibiotic treatments, providing more precise insight into their mechanisms of resistance. These preliminary results indicate the potential that this new window into intracellular bacterial communication provides, and motivate the continued exploration of models to unveil the mechanisms underlying AMR.