I felt totally comfortable using Qiime1 with the 16S SILVA database but now I’m totally confused using Qiime 2.
First of all I want to know if I have to use a trained classifier? In Qiime1 it was easy to change the database by using the .qiime_config file. Is it possible to do the same in Qiime2?
The second question is about the data import. I implemented a quality trimming pipeline for my data which I used before the split-libraries.sh script in Qiime1. After the quality trimming pipeline my paired-end fastq files are merged, trimmed, quality filtered and the primers are removed.
I still want to use that pipeline in combination with Qiime2. But how should I import my data now?
Short answer: no. We have two other classifiers in qiime2, classify-consensus-blast and classify-consensus-vsearch that operate directly on reference sequences/taxonomy (imported as QIIME2 artifacts), rather than using a trained classifier.
But the other way to think about this is that you pre-train your classifiers ahead of time and they become your reference databases, which you can switch at will when doing different analyses.
In QIIME2 there is no congif file. You would just need to point to different file paths (or you can define variable names that point to these file paths in your shell, so that you just switch the variable names to use different classifiers with your data without typing out a full file path)
Those data can then be used with otu picking in the q2-vsearch plugin, or denoising with deblur.
By re-assembling your processing pipeline in QIIME2, you will preserve all of that information in provenance to make it easier to trace precisely how a data file was generated (see a description of provenance here).
You can also use dada2, which will perform denoising and joining all together in one command, eliminating the need for the processing pipeline that you have been using. See this tutorial for an example of running dada2 on paired-end data.