Bayesian multinomial logistic-normal models have proven useful for the analysis of microbiome data. Yet the computational challenge of inferring thees models has prevented their widespread use. In this talk I will introduce multinomial logistic normal models for linear and non-linear regression as well as time-series models for the analysis of microbial time-series. I will show that all of these models can be inferred efficiently using a unified approach that has been implemented in the software package mongrel. I will show that inference using our approach is often five order of magnitude more efficient than optimized MCMC while maintaining accuracy of point estimation and uncertainty quantification. The methods I will present enable scalable inference in Bayesian multinomial logistic-normal models.