High NDR values when quantifying transposable elements in exosomes

[NaixiangYu]

Dear Bambu team,

Thank you for making such an amazing tool available!
I'm analyzing ONT-cDNA data of porcine exosomes with Bambu for downstream transposable element (TE) quantification, and consistently get a high recommended NDR (~0.685) with both the pre-trained human model and sample-specific training.

Could you please advise:

1. Should I set NDR = 1 and min.txScore.singleExon = 0 for downstream TE overlap analysis (e.g., with RepeatMasker)?
2. Would you recommend sticking with the human model in this case?

I would greatly appreciate your thoughts and suggestions.

Best regards,
Yu

Bambu warning for sample specific training:

Bambu warning for pre-trained human model:

[SuiYue-2308]

Hi,

Sorry for the late reply.

Setting NDR = 1 will report all transcripts, including those with low confidence, which I generally don’t recommend. However, for transposable elements (which often include single-exon transcripts), I suggest using min.txScore.singleExon = 0. You can then use the resulting txScore to manually evaluate their confidence.

Not sure about the differences between human and porcine annotations: I recommend using a porcine closely related annotation to train the model before inputting it into Bambu for the best results.

Best regards,
Yue