Error in Clustering mode: Invalid clustering mode: 3

[CWYuan08]

Hi, I am trying to run isONclust2 first for isONcorrect, but I got this error for all the batches, one example:
Loaded input batch from batches/isONbatch_9.cer:
Batch number: 9
Batch range: [244492,273799]
Depth: -1
Nr sequences: 29308
Nr bases: 50001212
Nr clusters: 29308
Nr nontrivial clusters: 0
Minimizers in database: 0
Created pseudo-batch for single clustering:
Batch number: -9
Batch range: [244492,273799]
Depth: -1
Nr sequences: 29308
Nr bases: 0
Nr clusters: 29308
Nr nontrivial clusters: 0
Minimizers in database: 0
Resetting input clusters.
Clustering mode: Invalid clustering mode: 3

from running:
for f in batches/isONbatch_.cer; do
filename=$(basename "$f")
output="clustered/${filename%.}.cer"
isONclust2 cluster -v -l "$f" -o "$output"
done

could you please advise what I need to fix?

Many thanks!!

Best,
CW

[ksahlin]

Hi @CWYuan08,

Since I was the one referring you here.. This seems to be the way to run isONclust2: https://github.com/epi2me-labs/wf-transcriptomes (and this section in particular: https://github.com/epi2me-labs/wf-isoforms#de-novo-based-approach-experimental)

You can then run isONcorrect on the clustered output, and isONform for consensus. I have not tried the approach they listed here, but they say it is experimental, which typically means no substantial benchmarks have been done.

Best,
K

[Johnsonzcode]

Hi @ksahlin
But de-novo-based-approach-experimental cannot be runned on command line mode. You can see here.
So maybe this pipeline isONclust-isONcorrect-isONform is the only way ?

[cjw85]

(Thanks @ksahlin for adding some comments here).

@Johnsonzcode the de-novo based approaches are indeed still largely experimental and so the code is not well-maintained. This project is not currently maintained and there is no one at Oxford Nanopore Technologies currently studying de-novo approaches. I dare say that @ksahlin is far more of an expert in the space than we are.

[Johnsonzcode]

So How could I get non-redundant isoform from ONT full-length transcripts. Is there some pipeline suggested ？

[Johnsonzcode]

(Thanks @ksahlin for adding some comments here).

@Johnsonzcode the de-novo based approaches are indeed still largely experimental and so the code is not well-maintained. This project is not currently maintained and there is no one at Oxford Nanopore Technologies currently studying de-novo approaches. I dare say that @ksahlin is far more of an expert in the space than we are.

But how could I sovle the error as mentioned? Or is there some pipeline suggested to get non-redundant isoform from ONT full-length transcriptome sequencing ？

[ksahlin]

Or is there some pipeline suggested to get non-redundant isoform from ONT full-length transcriptome sequencing ？

I can suggest running pychopper-isONclust-isONcorrect-isONform for this. The problen is that isONclust does not scale to very large datasets. This is what @CWYuan08 noticed and, hence, we ended up here looking for isONclust2 to replace isONclust as a solution. Another way is to manually batch (i.e. split) your large dataset to independent instances that isONclust can run on.

[Johnsonzcode]

Or is there some pipeline suggested to get non-redundant isoform from ONT full-length transcriptome sequencing ？

I can suggest running pychopper-isONclust-isONcorrect-isONform for this. The problen is that isONclust does not scale to very large datasets. This is what @CWYuan08 noticed and, hence, we ended up here looking for isONclust2 to replace isONclust as a solution. Another way is to manually batch (i.e. split) your large dataset to independent instances that isONclust can run on.

This pipeline may work.

[cjw85]

@Johnsonzcode

The https://github.com/epi2me-labs/wf-isoforms pipeline is deprecated and its functionality is folded in to wf-transcriptomes. If you wish to use the de-novo route through wf-transcriptomes, lets work to uncover the bug you are seeing with its use on the issue you have already started over there. I feel we've gone a bit off topic from @CWYuan08's original post here.

[CWYuan08]

Dear @Johnsonzcode, @cjw85, @ksahlin,
thank you all for the useful discussions here, this is what I would like to ask and follow too!
Best,
CW