bonito mod-calling with remora model

[hd2326]

Greetings!

I have a trained onnx remora model, and I am wondering whether would it be possible to convert it to the tar+toml format for bonito mod-calling. Thank you very much in advance for your help!

[iiSeymour]

Hello @hd2326 you can give bonito your custom trained remora onnx model with bonito basecaller --modified-base-model custom.onnx (it doesn't need converting).

[hd2326]

@iiSeymour Thank you very much for the quick reply!

So the mod-calling will be a two-step process, 1) using bonito tar+toml for basealling, based on which 2) using remora onnx to make mod-calling, right?

[iiSeymour]

You need both models (a bonito basecalling model [tar+toml] and a remora modbase [onnx] model) but it's one command -

bonito basecaller [email protected] /data/reads --modified-base-model custom.onnx > calls.bam

[hd2326]

Got it! Thank you so much for the explanation!

[hd2326]

Greetings!

As I am running bonito as @iiSeymour suggested:

bonito basecaller $bonito_models/[email protected]/ $rawdata --modified-base-model $remora_model/model_best.onnx --modified-bases $mod --reference $genome

I got the following error:

remora.RemoraError: No trained Remora models for /bonito_models/dna_r9.4.1_e8. Options: dna_r9.4.1_e8, dna_r9.4.1_e8.1, dna_r10.4_e8.1

It seems that the remora model I provided cannot be recognized. Any insights on the issue? Thank you very much!

[marcus1487]

The --modified-bases argument triggers bonito to lookup the corresponding remora model. But it appears that you have also specified the path to a remora model with the --modified-base-model argument which specified the modified bases to call. Removing the --modified-bases argument from the call should work.

[hd2326]

Awesome! As @marcus1487 suggested, removing --modified-bases solves the problem!

But another problem came...

It seems that the bonito bam files are not compatible with samtools mpileup for modification analysis. I got the samtools mpileup: error reading from input file error, but I don't have this problem for guppy bam files.

Specifically, the modification I am trying to analysis is uracil (I named it x and 5xT) in DNA, and I trained the bonito model using the following workflow:

1. I ran taiyaki prepare_mapped_reads.py with --mod x T 5xT to generate the hdf5 file.
2. I ran remora dataset prepare with --motif T 0 to convert the hdf5 file to the npz file.
3. I ran remora model train with the provided ConvLSTM_w_ref.py model template.

The MM tag I got in bonito bam files is like MM:Z:['T']+x,-1,...,-1;. As for guppy bam files I got something like Mm:Z:C+m,0,...,0;. Not sure what does the negative MM value mean, and maybe that causes the incompatible problem? Any insights on the issue? Thank you very much!

[najohink]

Hi @hd2326,

I was wondering if you ended up getting bonito to basecall Us?

best, S

[hd2326]

@najohink Actually no. Still the same error...