Support for R10 cDNA?

[jbeaulaurier]

Hello,

Thanks very much for this tool! Are you planning to add a model for R10 cDNA any time soon? I suspect that it would get a lot of use.

Thanks,
John

[zhengzhenxian]

Hi John,

To train a R10 cDNA model, we need at least one GIAB R10 cDNA sequencing dataset. Unfortunately, we have not been able to find any publicly available data for this purpose. Pls kindly let us know if you are aware of any sources we might have missed.

Many thanks.

Wishes,
Zhenxian

[tomoosting]

Sorry for opening this issue back up. As a follow-up question to that from John. Given that there is no specific model available for cDNA with the 10.4 chemistry. Which model would you recommend if you would like to run clair3-rna anyway? Or would you strongly advise against using clair3-rna.
If so, do you have any other recommendations for variant calling cDNA-PCR (SQK-PCS114) data generated on R10.4 flow cells?

Many thanks,
Tom

[aquaskyline]

we have a R10.4 cDNA model on the way, stay tuned. Tom Oosting ***@***.***> 于 2025年1月31日周五21:18写道：

…

Sorry for opening this issue back up. As a follow-up question to that from John. Given that there is no specific model available for cDNA with the 10.4 chemistry. Which model would you recommend if you would like to run clair3-rna anyway? Or would you strongly advise against using clair3-rna. If so, do you have any other recommendations for variant calling cDNA-PCR (SQK-PCS114) data generated on R10.4 flow cells? Many thanks, Tom — Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAG2SOLNI34DFHJ6PBQ6NPD2NNZ3BAVCNFSM6AAAAABNHDQEM2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMMRXGMZDEMRVGI> . You are receiving this because you modified the open/close state.Message ID: ***@***.***>

[youyupei]

Hi @aquaskyline , may I have a follow-up to Tom’s question. Before the formal release of the R10 model, would using the current ont_guppy_cdna model on R10 data called by Dorado be plausible, or should it be strictly avoided?

[aquaskyline]

Thank you for asking. Yes we have r10 cDNA downloaded so a model specifically for r10 cDNA data is just a matter of time. We have been occupied recently but we target to release the r10 cDNA by the end of March.

[zhengzhenxian]

Hi @tomoosting @youyupei @jbeaulaurier,

The R10 cDNA model is now available in our latest release (v0.2.2). To use it, you can pull our most recent Docker image and specify the --platform ont_r10_dorado_cdna option. Please note that this model was trained using the GIAB dataset basecalled with Dorado.

Kindly let us know for any issues with it, thanks!

[sparthib]

Hi @aquaskyline , can the R10 dorado cDNA model also be used for R10 directcDNA data basecalled with guppy?
Or would the R9 guppy cDNA model work better?

Thanks,
Sowmya

[aquaskyline]

ont_dorado_drna004 is for the latest dRNA sequencing already.

[sparthib]

Thanks for your quick response, @aquaskyline
Is the dRNA also compatible with direct-cDNA data or does it only work with direct RNA?

Thanks,
Sowmya

[aquaskyline]

They are totally different so use the correct model.

[sparthib]

Thanks for your response @aquaskyline. It looks like all the ONT pre-trained models are either made for direct RNA or PCR cDNA. Not direct-cDNA, is that right?

Do you have any suggestions for working with direct-cDNA instead?

Thanks,
Sowmya

[airichli]

Hi @tomoosting @youyupei @jbeaulaurier,

The R10 cDNA model is now available in our latest release (v0.2.2). To use it, you can pull our most recent Docker image and specify the --platform ont_r10_dorado_cdna option. Please note that this model was trained using the GIAB dataset basecalled with Dorado.

Kindly let us know for any issues with it, thanks!

Great! May I know the performance of the R10 cDNA model (e.g., Recall, precision, F1-score)? The R9 cDNA model seems still have some space to improve. Thanks.

[aquaskyline]

Thanks for your response @aquaskyline. It looks like all the ONT pre-trained models are either made for direct RNA or PCR cDNA. Not direct-cDNA, is that right?

Do you have any suggestions for working with direct-cDNA instead?

Thanks, Sowmya

We suggest using the PCR cDNA model and a turn-around. We don't have direct cDNA data for model training, but will look into it when we have it. Let us know if you can share with us some direct cDNA data of the GIAB reference samples.