Adapt pipelines to the COI fragment. Suitable adaptations include read processing and filtering steps that leverage evolutionary properties of the protein coding nature of this fragment, or determining appropriate parameters for tools originally designed for other DNA regions. Some recent advances have been made in filtering tasks (metaMATE, Andújar et al., 2021; coil, Nugent et al., 2020; entropy-based denoising, Turon et al., 2020) but further development in these promising areas is essential to fully exploit the potential of the COI gene for metabarcoding. As mentioned previously, there are no tools that enable simple length filtering variation that accounts for codon-level insertion or deletion. To our knowledge there is limited work exploring the extent to which protein structure inference might allow identification of erroneous sequences: for example the SOAPbarcode pipeline (Liu et al., 2013) includes a script that filters sequences based on translation hydrophilicity, but this is not comprehensively documented or discussed in the associated publications. Computation of protein structural properties is relatively trivial to perform, and seems like a fertile ground for novel development of filtering tools for protein coding markers.