ksahlin
commented
1 year ago Yes, there are several parameters to adjust (NGSpeciesID uses isONclust for the clustering step). However, NGSpeciesID is not built for exact separation/clustering of sequences at a pre-determined exact similarity rate.
You can mimic very stringent clustering by setting large --k, lower --w and high --mapped_threshold and --aligned_threshold. --mapped_threshold and --aligned_threshold could be set to 0.9 and --k between 20 and 30 and --w between k+10 and k+30 if you are working with sequences without much errors.
The above suggestion will not work when your sequences have higher error rates than around 2%, such as in ONT long reads.
FYI, all of these parameters relates to the clustering:
--k K Kmer size (default: 15)
--w W Window size (default: 50)
--min_shared MIN_SHARED
Minmum number of minimizers shared between read and cluster (default: 5)
--mapped_threshold MAPPED_THRESHOLD
Minmum mapped fraction of read to be included in cluster. The density of minimizers to classify a region as mapped depends on quality of the read. (default: 0.7)
--aligned_threshold ALIGNED_THRESHOLD
Minmum aligned fraction of read to be included in cluster. Aligned identity depends on the quality of the read. (default: 0.4)
--min_fraction MIN_FRACTION
Minmum fraction of minimizers shared compared to best hit, in order to continue mapping. (default: 0.8)
--min_prob_no_hits MIN_PROB_NO_HITS
Minimum probability for i consecutive minimizers to be different between read and representative and still considered as mapped region, under assumption that they come from the same transcript (depends on read quality).
(default: 0.1)
Is there a way to adjust clustering parameters?
for example, some OTU pipelines will generate different number of bins depending on whether you want 98% or 95% similarity clusters. I do not see options for this for NGSpID. on that note, what IS the perc. threshold used here?