Bioaugmentation is receiving increasing attention as a green technology to treat contaminated areas by inoculating specific biodegrading microorganisms. However, our understanding of the role of microbial community composition and structure in the expression of contaminant degradation potential is yet to improve. It could help making wise choice for microorganisms – community or specific strain – to be inoculated in contaminated soils with consideration to their indigeneous microbiota. Here we tried to predict the microbial degradation of two herbicides, glyphosate and isoproturon by means of penalized regression and machine learning methods routinely used in genomic selection. To this end, we conducted experimental modifications of these two herbicides degrading communities by applying biocide treatments coupled with serial dilutions. We then applied three selected genomic selection methods (i.e. Ridge Regression, LASSO and Random Forest) on these community variants to link their OTUs composition to their herbicide degradation capacities. Resulting predictions power is compelling with more than 80% correlation between predicted and actual herbicide degradation capacities. Moreover, OTUs detected as having an impact on herbicide degradation were confronted to literature and validated. Finally, futures applications in a bioremediation perspective are considered.