COI analysis

Set marker_gene: COI in the parameters file.

ASVs or OTUs

Set clustering_algo to swarm (ASVs) or vsearch (OTUs); see 16/18S analysis for both.

Hint
The differences (d) parameter of Swarm can take rather high values for COI data: it has been shown that 10 < d < 25 can work well. Fastidious mode only applies to d = 1, so PEMA turns it off otherwise.

Taxonomy assignment

RDPClassifier with MIDORI

With classifier: rdp, PEMA assigns the taxonomy with RDPClassifier, trained on MIDORI:

• midori_1: MIDORI 1, metazoans;
• midori_2: MIDORI 2, all eukaryotes; about 100,000 more species, but considerably slower and needing more memory.

Set database to one of them and databases_dir to a folder of yours (in the rdp section): PEMA downloads the database from Zenodo the first time it is used, and reuses it afterwards. Set memory to at least 20 GB; with MIDORI 2 you will most likely need an HPC system or a server (see Running on HPC).

RDPClassifier gives a confidence to every rank; PEMA keeps the ranks from the top down while their confidence is at least confidence (default 0.8; for short amplicons a lower value, such as 0.5, is common).

crest4 with MIDORI

Alternatively, with classifier: crest and database: midori253darn, the crest4 MIDORI database is used (see 16/18S analysis for crest4).

Your own reference database

Both classifiers can be trained on your own reference sequences; see Training the RDPClassifier and Training the CREST classifier.