Guppy basecaller ONT

Question

[samuel]

Hi, I'm a complete beginner at ONT. I've read that you can use Guppy for basecalling for the Oxford Nanopore Technology. Could someone point me to a tutorial/reference manual/ test data??? I can't seem to find much information on this anywhere???

[GenoMax]

@Wouter de Coster (biostars ONT expert) should have something more detailed for you tomorrow but in the mean time take a look this post to find the links to download the basecaller from. I hope you have a compatible GPU to run it on.

[WouterDeCoster]

Guppy also works on CPUs, although indeed much faster on a GPU. But that may not matter that much if you're not processing tons of data.

[robert.muldowney]

Is there a list of compatible GPUs? I have a NVIDIA K40c and Quadro K2200 and both seem to be incompatible as I get an error message when I try to use them.

Loading fatbin file shared.fatbin failed with: CUDA error at /builds/ofan/ont_core_cpp/ont_core/common/CUDAHelper.cpp:40: CUDA_ERROR_NO_BINARY_FOR_GPU Failed to load shared(shared) from fatbin CUDA runtime compilation not supported in this build. Segmentation fault (core dumped)

[WouterDeCoster]

quoting from the community pages:

The GPU-enabled ont-guppy package will not install a GPU driver by default – it will be necessary to install this yourself e.g. by visiting the NVIDIA website. Guppy requires an NVIDIA driver of at least version 384.

and also

The Guppy CUDA kernels are compiled for compute version 6 and higher, and anything lower than that will not work out of the box. You can check your NVIDIA GPU's compute version here.

[andorjkiss]

Hi,

If you set-up your NVIDIA drivers and CUDA toolkit properly, you just have to download the GUPPY for GPU tarball (64-bit) and then decompress the tarball, and execute the basecaller:

./ont-guppy/bin/guppy_basecaller -i input_dir -r -s output_dir/ -c dna_r9.4.1_450bps_hac.cfg -x 'cuda:0 cuda:1' --compress_fastq

This worked very well for me on dual NVLINKed RTX TITANs on Ubuntu 18.04.3 LTS with the nVIDIA 440 driver and the CUDA-10.2 toolkit.

[WouterDeCoster]

Is that the most recent version, 3? I'll see if I can find a list tomorrow.

[BioinformaticsLad]

Is he putting out a tutorial on basecalling?

You can check this out for details on guppy parameters.

[WouterDeCoster]

Is he putting out a tutorial on basecalling?

According to my sources, no :-D

[BioinformaticsLad]

Wouter de Coster (biostars ONT expert) should have something more detailed for you tomorrow

What did genomax mean by this? I thought he might be referring to a tutorial or paper.

[WouterDeCoster]

Genomax was essentially creating high expectations. But I don't think any tutorial or paper I could write would improve on the official documentation, so I won't.

[GenoMax]

@BioinformaticsLad: Problem is that the community you posted a link to is not open. It requires an invite/purchase from ONT to join.

[WouterDeCoster]

Yes, I have raised this issue in the ONT community. I would have to look into terms and conditions to check what I can share outside of the community.

[samuel]

I could access the link - thank you. Could you clarify something for me? I was just told that you can either get base calls directly from the grid/minion (via Guppy) or use the fast5 files to base call with Guppy separately. Is this correct? If so, is there an advantage/disadvantage to base calling separately??

[BioinformaticsLad]

That's right. You can basecall the fast5 files separately (from sequencing).

Advantages of separating the processes: 1) You're sequencing on a minION and have a weak PC. Simultaneously basecalling can slow things down. You can move your fast5 files to a much faster PC for more efficient basecalling. 2) You want to basecall using something other than Guppy. 3) You want to re-basecall your old fast5 files using an updated & improved basecaller (ONT updates their software regularly giving incremental increases in accuracy).

Disadvantages: 1) No real-time data stream of reads and/or real-time analysis (eg. EPI2ME) 2) If you have a GridION, basecalling is fast enough to keep up with sequencing so separating the processes will take you longer to get to reads. 3) You can't see the quality of your reads during the run.If you could, you might decide to stop the run due to low quality reads. 4) I've heard that (on Promethion at least) using all flowcells simultaneously sequencing and basecalling can slow things down.

[samuel]

Thanks for that - most informative and very much appreciated.

Answer 1

[lakhujanivijay]

I agree to genomax 200 % ; WouterDeCoster is the best person to answer!

However, if you have access to nanopore community page then this link is the best place to look at

[WouterDeCoster]

That link would definitely answer most, if not all questions. Can you access it zoegward ? Let us know if there is anything more specific you struggle with.

[lakhujanivijay]

WouterDeCoster, that link is dead

[WouterDeCoster]

I was talking about your link. My link was just tagging OP, but it seems the trailing ? corrupted it :) (changed it now)

[lakhujanivijay]

strange, i still see this

[WouterDeCoster]

Oh yes, there is some weird issue with OPs user profile. I forgot. I already informed Istvan Albert

[tania.gonzalez]

Hi, I'm a bioinformatics student and am trying to analyze a data generated with a MinION flowcell (9.4.1D). I am new to Nanopore data analysis and was wondering how could I get access to the ONT community webpage for more information on guppy parameters and how to analyze my data.

[GenoMax]

AFAIK someone still needs to invite you to join the community. It can be the person who originally bought the minION.