The COI locus differs from many other metabarcoding loci (e.g. 18S, 16S, 12S, ITS) in that it is a protein coding gene, imparting strict expectations of amplicon sequence read properties that can be exploited in metabarcoding bioinformatics (Andújar et al., 2018b). However, the adaptation of pipelines to this fragment are in general rarely implemented in the papers of the core set. For example, only 22 papers (20%) used amino acid translations to identify erroneous sequences (“translation filtering”),  using 11 different software tools for the task. The reason for low implementation of translation filtering is likely twofold; first, none of the major metabarcoding software packages include functions for translation filtering, and second, there is no standard straightforward command line software for undertaking this task. Those papers that carry out translation filtering do so by using one of three main approaches: (i) sequences are viewed and translated in a GUI application such as Geneious (https://www.geneious.com) or MEGA (Kumar et al., 2018), and those with stop codons manually removed, (ii) sequences are processed through a custom script, some of which are available on github but none of which are used by research groups separate from the author, and (iii) sequences are aligned against references using MACSE (Ranwez et al., 2011) and those containing indels or stop codons are  removed. The first option is time consuming and prone to human error, and custom scripts are challenging to document and maintain for a wider number of users. While MACSE is the most frequent single approach, it is computationally efficient only for small datasets. There may be some potential in the recent coil R package (Nugent et al., 2020) that uses Hidden Markov Models to identify and filter translation-based errors and appears to scale well to large datasets, although the R implementation presents a slight barrier to efficient inclusion in pipelines. Furthermore, the majority of translation filtering approaches are based solely on removing stop codons, while there may be other potential avenues for filtering based on amino acid translation. The extent to which expectations for protein structural properties can be applied to metabarcoding sequences for filtering other non-synonymous errors has been underexplored (but see Turon et al., 2020).