Deciphering the microbial architecture of pesticide and antibiotics biodegradation

Understanding emerging functions at the scale of a bacterial community is a major challenge in microbial ecology and could lead up to promising tools for engineering microbial communities, for example in bioremediation. Here, through a top-down approach we obtained compositional variants of pesticide and antibiotics-degrading communities and further investigated communities features associated with their degradation abilities. We first tested whether diversity index or functional genes abundance could reliably be used as a proxy for this function, and obtained encouraging, albeit variable results. Further, through the use of statistical tools borrowed from the genomic selection literature, we were able to derive accurate prediction of the mineralisation potential of a bacterial community, based on its composition. However, the parallel between genotype-phenotype and community composition-mineralisation potential suffers a crucial caveat: bacterial abundances vary on a much wider scale than allele dosage at a given locus and are prone to change over time (particularly at the mineralisation scale). Here we observed that using presence/absence data instead of relative abundance can overcome these limitations and provide a clearer functional signal for mineralisation prediction through linear regression models. Random forest can also intrinsically deal with microbial data without transformation and select for significant predictors. We suggest drawing inspiration from the tools and concepts used in genotype-phenotype mapping to elucidate microbial functions at the community level while keeping in mind the significant differences between these two fields. This parallel is here exemplified by the concept of microbial architecture of degrading functions, akin to the genetic architecture of phenotypic traits.