The hybrid open-reference method steers a middle course: Since data that are not represented in the reference are recovered, bias driven by differential representation in the reference is reduced. In addition, since the amount of data being fed into the de novo step is minimized, the impact of study-specific error profiles is diminished. Open-reference OTU picking is modestly parallelizable and can be augmented with techniques such as use of a random subsample when constructing the intermediate de novo reference in order to accelerate its performance (details on the procedure can be found in [45] but are generally handled in by software without end-user intervention). However, open-reference picking shares a number of drawbacks with the de novo strategy, including necessity for pre-filtering, unsuitability for data from multiple variable regions, and necessity for re-picking when combining studies together. Of course, given the continued expansion of computational resources, sequencing throughput, and completed genomes available, optimal strategies for overcoming technical hurdles and enabling meta-analysis will require ongoing re-assessment.
