With the advent of 16s rRNA high throughput technology, there has been considerable interest in studying changes in the human microbial composition with respect to ordered explanatory variables, such as concentration of a toxin, ordered disease groups (e.g. disease free, mildly sick, sick, very sick) etc. For example, a toxicologist may be interested in understanding how the gut microbiome composition varies with different concentrations of a toxin. They may be interested in identifying taxa whose relative abundances increase with toxin levels and those that decrease. In this talk, we introduce a novel methodology, inspired by order-restricted inference, to cluster taxa with similar dose response patterns, where the patterns are represented by mathematical inequalities. The proposed methodology takes into account the simplex structure as well as excess zeros in the data. We illustrate the methodology using a Norwegian oral microbiome data, where subjects are exposed to a wide range of chemicals.