ksahlin
commented
4 years ago Hi @nextgenusfs, thank you appreciate it!
Regarding the runtime bug: I actually fixed it today (see #6). So you can either reinstall IsoCon (version 0.3.3) by removing and reinstall from scratch, or it is simply sufficient to downgrade networkx to 2.3 as follows:
pip uninstall networkx
pip install networkx==2.3 As for the strategy I will get back tomorrow (it's pretty late in my timezone). Here are some general comments on IsoCon for ONT sequencing:
- IsoCon does not handle reverse complements. If you have reverse complemented sequences the predictions will contain the sequence and its reverse complement, but maybe that’s easy to post-filter? Another strategy is to identify the primers beforehand and re-orient the reverse complements (to speed up runtime of IsoCon even more).
- There will be some redundant consensus due to the different ONT error profile compared to IsoSeq data. Therefore, parameters to specify would be
--max_phred_q_trusted 20(default is 43 for hihger quality CCS reads) and--p_value_threshold 0.00001(instead of default 0.01). This could however also be post-filtered by simply removing consensus with a p-value larger than e.g., 0.00001 (the p-value is printed to the accession of the consensus sequence)
nextgenusfs
Hi @ksahlin. Thanks a ton for all of your tools with noisy reads. I'm looking for a solution for de novo clustering of ONT amplicon reads from environmental sequencing, ie fungal rRNA amplicons. The data I'm trying this on is from a mock community of mixed species. The region is the ITS-LSU region of rRNA in fungi -- we typically define species with a 97% pident cutoff with this region. The data has been pre-processed by re-orienting reads into the same direction and finding/trimming both forward and reverse primer sequences.
I've tried isONclust and at first it seemed like it might be working great (and quite fast), but then on further inspection it was a little too liberal on clustering the data that I have access to at the moment, effectively combining too many reads into the same "gene family". I ran a parameter search by varying k and w to see if I could get it to give me the proper results, but essentially never got a set of parameters that could delineate the clusters properly. My goal is to find a method to identify the "centroid", as then it is relatively straightforward to use
spoaandracon/medakafor error correction. I tried to clean up the clustering little bit by invoking a "sub clustering" by plotting read lengths (as fungal ITS-LSU sequences are variable in length) and then pulling out "peaks" from the lengths of reads -- this seemed to work okay, but still not quite what I'm looking for.Based on some of the other issues in your tool repositories, I then tried
IsoConwhich you had indicated seemed to be a more general approach.IsoConhas a much much longer runtime and then eventually crashed with the error below (note I ran it initially without--prefilter_candidates --min_candidate_support 2and it crashed with same error).If you have any other suggestions on an appropriate workflow I'd be grateful to hear your opinions.
Thanks, Jon