IgBLAST update in Docker image?

[wyattmcdonnell]

Hi Mike et al.!

Just wondering if there are plans to update the Docker image to IgBLAST 1.8? We're seeing some differences in performance/reconstruction between our local version of the TraCeR pipeline which runs on IgBLAST 1.8 and the Docker image which runs off of IgBLAST 1.7.

[mstubb]

This should definitely be possible. First it'd be really interesting to know what differences you're seeing between the two IgBLAST versions. Would you mind sharing some examples?

Thanks!

Mike

[wyattmcdonnell]

Awesome, thanks Mike!

Happy to share an example—in this instance, we're seeing the same alignments against TRBV6-2 and TRBV5-1 with both IgBLAST 1.7 and IgBLAST 1.8, but TraCeR will only reconstruct contigs with the alignments from IgBLAST 1.8.

We're running the same command pipe between versions, which looks like this:

tracer assemble --loci A B G D --ncores $ncores --species Hsap $FASTQ1 $FASTQ2 $cell_name tracer_out

It doesn't seem to make a difference whether we use --m assembly or --m imgt, nor if we use the --small_index flag.

The paired end files are a bit big so here's a Dropbox link to the FASTQs for the T cell described.

On an unrelated note, running the final FASTAs from TraCeR through IMGT/V-QUEST (or HighV-QUEST) and searching for insertions/deletions can also retrieve slightly different results for a small number of cells (out of 679 successful reconstructions, insertions/deletions affecting productivity or alignment detected in ~1-2%).

Talk soon,
Wyatt

[mstubb]

Hi Wyatt, Thanks for this. I just ran TraCeR using the FASTQs from Dropbox as input. There was a *very* low alignment rate so I couldn’t proceed with testing. Only three reads aligned to D and one read aligned to G. How were these files generated? I presume they’re not the ones that you originally used for input? Would you be able to make the original input fastqs available via FTP or similar so that I can do some testing? Otherwise, please could you zip the tracer output directories for this file from both of your versions of tracer and send those to me? Thanks a lot, Mike

…

On 17 Apr 2018, at 21:45, wyattmcdonnell ***@***.***> wrote: Awesome, thanks Mike! Happy to share an example—in this instance, we're seeing the same alignments against TRBV6-2 and TRBV5-1 with both IgBLAST 1.7 and IgBLAST 1.8, but TraCeR will only reconstruct contigs with the alignments from IgBLAST 1.8. We're running the same command pipe between versions, which looks like this: tracer assemble --loci A B G D --ncores $ncores --species Hsap $FASTQ1 $FASTQ2 $cell_name tracer_out It doesn't seem to make a difference whether we use --m assembly or --m imgt, nor if we use the --small_index flag. The paired end files are a bit big so here's a Dropbox link to the FASTQs <https://www.dropbox.com/sh/cknysg62ohqy2rk/AADP47T6TfII1NMNeTnHs8ZJa?dl=0> for the T cell described. On an unrelated note, running the final FASTAs from TraCeR through IMGT/V-QUEST (or HighV-QUEST) and searching for insertions/deletions can also retrieve slightly different results for a small number of cells (out of 679 successful reconstructions, insertions/deletions affecting productivity or alignment detected in ~1-2%). Talk soon, Wyatt — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#67 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABFwhnsWJ2ZpcG_j3JvkQlD3izwqoIl0ks5tpsUEgaJpZM4TXno9>.

[wyattmcdonnell]

Hey Mike!

Let me work on zipping those results up. Note that these are SmartSeq v2 libraries that were sequenced on a NovaSeq, so the percentage alignment will likely be low for most of the cells—but that number of aligned reads is low enough to make me wonder if I shared the wrong file.

These are the input FASTQs we used, and we're still getting a very good recovery rate with TraCeR running IgBLAST 1.8 (and lower with IgBLAST 1.7 as mentioned above).

Thanks,
Wyatt

[mstubb]

Hi Wyatt, Yes, I’d be suspicious that those are not the correct files. There isn’t enough alignment to provide any assemblies. Anyway, looking forward to seeing those zipped results. M

…

On 18 Apr 2018, at 22:04, wyattmcdonnell ***@***.***> wrote: Hey Mike! Let me work on zipping those results up. Note that these are SmartSeq v2 libraries that were sequenced on a NovaSeq, so the percentage alignment will likely be low for most of the cells—but that number of aligned reads is low enough to make me wonder if I shared the wrong file. These are the input FASTQs we used, and we're still getting a very good recovery rate with TraCeR running IgBLAST 1.8 (and lower with IgBLAST 1.7 as mentioned above). Thanks, Wyatt — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#67 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABFwhimINaC2mR6VQ94vrkpe1ft9Ivyeks5tqBrAgaJpZM4TXno9>.

[wyattmcdonnell]

Hey Mike!

I've sent an invite to your Wellcome Sanger Trust email (I know you're moving to 10X shortly if you haven't already!) to share these files via Box. This link is unique and only accessible in combination with your email. Please let me know if you can't access these; if not I'll make a unique Dropbox link for you.

Wyatt

[mstubb]

Hey Wyatt, Thanks for this. My Sanger email isn’t working any more now that I’ve left. Please can you re-share to mjt.stubbington [at] gmail.com. Thanks! Mike

…

On 20 Apr 2018, at 10:14, wyattmcdonnell ***@***.***> wrote: Hey Mike! I've sent an invite to your Wellcome Sanger Trust email (I know you're moving to 10X shortly if you haven't already!) to share these files via Box. This link is unique and only accessible in combination with your email <https://vanderbilt.box.com/s/3ee9e02qgqrgju4q5pdqg3690p8xuzuv>. Please let me know if you can't access these; if not I'll make a unique Dropbox link for you. Wyatt — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#67 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABFwhg9S-iKZYBiD_G-SiB7hj7vmvveKks5tqheJgaJpZM4TXno9>.

[wyattmcdonnell]

No problem! Just sent to your Gmail!

Wyatt

[mstubb]

Hi Wyatt, Thanks for this. I’ve just run IgBLAST 1.7 and 1.8 against the reconstructed TCRB sequences for cell 943-RG-7-i1_S204 (assuming that this is the one where you’re seeing differences) in the data that you shared with me. I get exactly the same VDJ calls from both versions. So, I think the differences that you’re seeing may arise from something earlier in the TraCeR pipeline. Please could you send me the output directory that you get just for that one cell from the Docker version as well as the non-docker version that you’ve run. If you have chance, please compare the versions of Trinity and Bowtie2 between the Docker and non-docker versions that you’ve used. It’d be great if you could also check the input files for that particular cell and send those to me if they’re different to the ones you sent before. Hope that all makes sense. Very best, Mike

…

On 20 Apr 2018, at 10:22, wyattmcdonnell ***@***.***> wrote: No problem! Just sent to your Gmail! Wyatt — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#67 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABFwhlool95QOS3XGcER3pLiVuChZ9Auks5tqhlpgaJpZM4TXno9>.

[wyattmcdonnell]

Hi Mike!

Sorry for the delay! After poking around in the Docker image, I think you're right—my guess is that this is due to a newer version of Trinity being used. Here's the versions we're using on our end for the non-IgBLAST tools:

Bowtie2: v 2.3.0
kallisto: v 0.44.0
trinity: v 2.6.6

As you suggested it seems likely that these changes are leading to the recovery of the TCR from those cells. All of the input files I sent you are correct—I should note that we are currently leaving the trimming step on just in case. Given the version differences, if you're willing to run a few cells on your end and see if things look any different that would be most helpful (I completely understand if you don't have time to do so). Do you have any concerns about using newer versions of these tools with TraCeR?

Thanks again,
Wyatt