Abstract
Motivation There are few tools that allow longitudinal analysis of metagenomic data subjected to distinct perturbations.
Methods This study examines longitudinal metagenomics data modelled as a Markov Decision Process (MDP). Given an external perturbation, the MDP predicts the next microbiome state in a temporal sequence, selected from a finite set of possible microbiome states.
Results We examined three distinct datasets to demonstrate this approach. An MDP created for a vaginal microbiome time series generates a variety of behaviour policies. For example, that moving from a state associated with bacterial vaginosis to a healthier one, requires avoiding perturbations such as lubricant, sex toys, tampons and anal sex. The flexibility of our proposal is verified after applying MDPs to human gut and chick gut microbiomes, taking nutritional intakes, or salmonella and probiotic treatments, respectively, as perturbations. In the latter case, MDPs provided a quantitative explanation for why salmonella vaccine accelerates microbiome maturation in chicks. This novel analytical approach has applications in, for example, medicine where the MDP could suggest the sequence of perturbations (e.g. clinical interventions) to apply to follow the best path from any given starting state, to a desired (healthy) state, avoiding strongly negative states.
