This will correct for effective reference lengths as done by almost any good tool those days. So the returned counts are not correlated with feature lengths. By default an expectation maximization algorithm is used to resolve multiple mappings of one read to many references which pretty much always happens in metagenomics data sets. The optimized likelihood function is very similar to the one in kallisto (https://doi.org/10.1038/nbt.3519).
count_references(object, ...)A data frame containing alignment (BAM) files, like generated by
find_alignments. Should contain columns "id", and "alignment".
A configuration as generated by config_count.
A list containing the used alignments and the transcript counts in `counts`.
Note that for the EM method there will be a NA reference reported which corresponds to the approximate abundance of references not contained in the database.