Request: Reenabling the no-trim parameter

[e-fuhrmann]

Question/Feature request

Please describe the issue:

Hello ONT-team!
I have a questions/request regarding RNA004 and its trimming.

Up until version 0.7.3 it was possible to use the no-trim parameter to disable adapter trimming for RNA. Ever since version 0.7.4/0.8.0 (4a28d58 if I'm not mistaken) this sadly seems to be no longer possible (as in, no-trim still exists, but has no effect).
This is actually a major problem for us, because we use protocols with custom sequence adapters. And with dorados automated trimming we observe significant overtrimming for them.
Previously, that wasn't a big deal as we just didn't trim and dealt with the adapters in downstream analyses. But now this is no longer possible (with newer versions), which leaves us in a bit of a pickle.

This leads me to my question/request:
Is there any other way to let dorado basecall (with newer versions) without the automated trimming?
Or, could you reenable the no-trim option for RNA adapter trimming?
Out of interest, could you elaborate on why this option was deactivate in the first place?

Thank you very much for your time!

Cheerio
~E

[malton-ont]

Hi @e-fuhrmann,

The reason we set dorado to always trim adapters for the RNA models is that our adapters are DNA and therefore can't be correctly called by the RNA model. Attempting this results in bad sequence that provides no useful information and, since the bad sequence also won't be found by the sequence-based trimming, we took the decision to remove it entirely.

There is actually a flag that will prevent removing the adapters via signal-space - --rna-adapters will make dorado think it can correctly basecall the adapter, so it will not remove it from the signal. You may still want to include --no-trim as well, as there will still be an attempt to find and trim an adapter in sequence-space.

If you're willing to share some data, we'd be interested in attempting to improve the adapter trimming to resolve the over-trimming you are seeing as well.

[billytcl]

Is the DNA adapter signal is so poor that it essentially gives random noise? Or is it systematic and can possibly be identifiable (even its garbled)?

[malton-ont]

@billytcl,

I wouldn't say the signal is "poor", exactly. It's just sufficiently different from RNA signal that the model is unable to interpret it correctly - since the RNA model hasn't been trained on that kind of data it won't give a reliable output for it. We actually rely on this difference in signal to locate the adapter for trimming, which is why I'd be interested to get a look at the data from @e-fuhrmann.

[e-fuhrmann]

Hi @malton-ont,

thanks a lot for the reply! I have tried the --rna-adapters parameter and can confirm that it works as you described!
Sadly, I cannot share any data regarding this at the current moment.

Cheerio
~E

[malton-ont]

@e-fuhrmann,

I'm glad that helped. Please be aware that this parameter is experimental, and it is possible that its behaviour may change in the future.