Dental caries or tooth decay is a prevalent global disease whose causative agent is the oral biofilm known as plaque. According to the ecological plaque hypothesis, this biofilm becomes pathogenic when external challenges drive it towards a state with a high proportion of acid-producing bacteria. Determining which factors control biofilm composition is therefore desirable when developing novel clinical treatments to combat caries, but is also challenging due to the system complexity and the existence of multiple bacterial species performing similar functions. Here we employ agent-based mathematical modelling to simulate a biofilm consisting of two competing, distinct types of bacterial populations, each parameterised by their nutrient uptake and aciduricity, periodically subjected to an acid challenge resulting from the metabolism of dietary carbohydrates. It was found that one population was progressively eliminated from the system to give either a benign or a pathogenic biofilm, with a tipping point between these two fates depending on a multiplicity of factors relating to microbial physiology and biofilm geometry. Parameter sensitivity was quantified by individually varying the model parameters against putative experimental measures, suggesting non-lethal interventions that can favourably modulate biofilm composition. We discuss how the same parameter sensitivity data can be used to guide the design of validation experiments, and argue for the benefits of in silico modelling in providing an additional predictive capability upstream from in vitro experiments.