ABSTRACT Microbiomes have been the focus of a substantial research effort in the last decades. The composition of microbial populations is normally determined by comparing DNA sequences sampled from those populations with the sequences stored in genomic databases. Therefore, the amount of information available in databanks should be expected to constrain the accuracy of microbiome analyses. Albeit normally ignored in microbiome studies, this constraint could severely compromise the reliability of microbiome data. To test this hypothesis, we generated virtualomes, virtual bacterial populations that exhibit the ecological structure of real-world microbiomes. Confronting the analyses of virtualomes with their original composition revealed critical issues in the current approach to characterizing microbiomes, issues that were empirically confirmed by analyzing the microbiome of Galleria mellonella larvae. To reduce the uncertainty of microbiome data, the effort in the field must be channeled towards significantly increasing the amount of available genomic information.

The virtualome: a computational framework to evaluate microbiome analyses