In contrast to closed-reference, a de novo approach consumes more computational resources but requires no pre-existing reference and allows a researcher to assign OTUs to as much of the data as possible, including OTUs never before observed. It is capable of producing phylogenies sensitive to subtle differences in OTUs (as the representative member for an OTU is an actual study sequence), but this very sensitivity means that contamination in the data (e.g., non-16S sequence such as phiX) will also be clustered into OTUs unless the contamination is explicitly filtered out prior to OTU picking and that the method is not suitable for data drawn from multiple variable regions (as highlighted in [31]). Additionally, the distinct error profiles of the studies being combined (which can stem from the 16S protocol, variation in the master PCR mix, error profiles of the sequencing instrument used, etc.) may lead to spurious, study-specific OTUs (for example, the GC bias of the Illumina platform can lead to sequences that contain more GC than another platform which can result in OTUs specific to the platform even if
