GROUP: Expanding the collection of iSEE modes

[csoneson]

The iSEE Bioconductor package enables interactive exploration of any type of data that can be stored in a SummarizedExperiment object. The default user interface of iSEE consists of a set of up to eight panels, each displaying a different aspect of the provided object (see e.g. http://shiny.imbei.uni-mainz.de:3838/iSEE/ for an example). The user is, however, free to change this interface interactively, by removing, adding or resizing existing panels, as well as by adding custom panels. In addition to changing the interface layout interactively, the user can pre-configure the set of panels, and even their content, and provide this information to iSEE when launching the application. In an attempt to simplify this procedure for commonly used setups, we implemented a few different iSEE "modes" (https://github.com/csoneson/iSEE/blob/master/R/modes.R), each of which defines a set of panels and content for the resulting application.

The purpose of this "hackathon" session is to expand the set of iSEE modes to cover a wider range of common applications. The expected outcome is a collection of new modes suitable for various types of data analyses. Depending on the participants, we will start by introducing the iSEE package and its capabilities.

Some suggestions for new modes to implement:

• scRNA-seq QC
• Single panel type applications
• bulk RNA-seq DE exploration

Topic leaders: Charlotte Soneson (@csoneson) and Federico Marini (@federicomarini) Scribes: Charlotte Soneson and Federico Marini

If you are interested in this topic, please give a thumbs up and/or tag yourself. If you have additional suggestions on what could be covered, please comment below.

[lgatto]

I would be interested in considering a proteomics mode. :-)

[federicomarini]

@lgatto can you suggest us a nice dataset (maybe on EH?) to start toying around with?

[mbstadler]

I would be interested in a mode that compares all available reduced dimensions in a single cell experiment

[federicomarini]

As a starting point for contributing, we did put together some resources.

The basics are included in https://github.com/csoneson/iSEEu - for which the vignette is rendered at https://csoneson.github.io/iSEEu/articles/iSEEu_contribute_eurobioc2019.html

Charlotte and I will open the session with a presentation which will also end up in the repo mentioned above.

We also assembled a small form for getting some extra feedback in case you cannot make it to the Developer Day/you prefer your comment to be done anonymously (you can still put your name in it if you want). You can find it here: https://forms.gle/Gv7uAouSKx92DrJ49

Looking forward to hearing your ideas and seeing them implemented together!

[kevinrue]

@mbstadler : to get you started, feel free to borrow from https://github.com/kevinrue/iSEE_custom/tree/master/plot_multiRedDim , it could use some improvement to generalize better with varying numbers of available reduced dimension results.

EDIT I just remembered that the session is about iSEE modes, while I linked to a custom panel. That said, it would probably be better to have a new mode that opens one Reduced dimension panel panel for each available reduced dimension, instead of stashing them all in a single custom panel. Bonus point if all the RedDimPlot panels are immediately linked to receive their selection from another panel.

[mbstadler]

That's exactly what I had in mind :-)

[lgatto]

Here's the proteomics usecase I was referring to:

On the left is a PCA plot at the protein level and on the right is the PCA plot for the peptides. The red did on the protein PCA plot corresponds to A2A791 and on the right, its peptides eVHEELAk, iIGDASTQTDALk, LFATkPELLDYk, NAkDEQGDLk, qFDQVSVFk and SFcSEGck.

The link mapping between peptides and their proteins is given by the ProteinDescriptions in the peptides' rowData. The data is available here.

[federicomarini]

A huge thanks to all participants!

Briefly wrapping up what came/will come out soon of here:

• Michael @mbstadler worked on the proposed multi-reduced dimension plot, steered by an additional panel for further exploration
• Pierre-Luc (@plger) is contributing a gene-structure centered panel - in the discussion later on, @jorainer suggested to leverage the EnsDb package family (in case this is not already the case 😃 )
• Fiona (@fionarhuang) completed a proof of principle for a custom panel for her TreeSummarizedExperiment package visualization
• Pablo (could not find a Github handle) worked on an exploration-oriented set of panels for digging deeper a set of candidate marker genes in a first round of redDim + featureAssay checks
• Charlotte (@csoneson) got a neat prototype running for the proteomics use case above, with a custom matched red dim plot for peptides and proteins
• I did a preliminary implementation of a panel set for what would be needed to explore bulk RNA-seq results (coming e.g. from DESeq2), with something that could resemble a layout for an app to pass over to wet-lab collaborators - a screenshot of that is here:

Again: Thanks to everyone! Do not forget to propose yourselves as contributors once it is ready to merge, and please do prettify any example or mini-vignette to help others understand what your mode is doing.

[jorainer]

If you need any help regarding the EnsDbs - just let me know. Happy to give a hand if needed.

[lgatto]

Is there a place/branch to follow (and possibly contribute) to the reddim module?

[federicomarini]

Yep: https://github.com/csoneson/iSEEu/tree/modeReducedDim

[mbstadler]

Thanks to all and especially Federico and Charlotte - I very much enjoyed the iSEEu hackaton. Best wishes, Michael

Sent from mobile device

---

Von: Federico Marini notifications@github.com Gesendet: Mittwoch, Dezember 11, 2019 9:08 PM An: Bioconductor/EuroBioc2019 Cc: Michael Stadler; Mention Betreff: Re: [Bioconductor/EuroBioc2019] GROUP: Expanding the collection of iSEE modes (#4)

Yep: https://github.com/csoneson/iSEEu/tree/modeReducedDim

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/Bioconductor/EuroBioc2019/issues/4?email_source=notifications&email_token=ADIGJ7SMPIIAZHYINZJB73TQYFCCHA5CNFSM4JOLYGH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEGUM6HQ#issuecomment-564711198, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ADIGJ7W4SEXLBSQVPEWRK53QYFCCHANCNFSM4JOLYGHQ.

[kevinrue]

Well done everyone! Sad that I couldn't make it at this European meeting, it sounded fun as always!

Can't wait to see the range of modes and community-contributed custom panels increase over time! I can tell you that iSEE-core is having a hackathon of our own to make that even easier in the near future!

[plger]

Ah, that's a much better idea than my rest thing! I'll put the code in a repo tomorrow and invite you if you'd like.Pierre-LucOn Dec 11, 2019 8:00 PM, Johannes Rainer notifications@github.com wrote:If you need any help regarding the EnsDbs - just let me know. Happy to give a hand if needed.

—You are receiving this because you were mentioned.Reply to this email directly, view it on GitHub, or unsubscribe.

[jorainer]

Sure @plger!

[federicomarini]

@plger Feel free to open up a branch directly here if you want 😉

[plger]

for the moment I've put it here: https://github.com/plger/iSEE_gviz/ since it's not yet in the format that iSEE_custom expects and there are a few things I still wanted to do... essentially Charlotte's with a few more options, and looking up ensDb (@jorainer any feedback on how I've done it would be appreciated!) rather than providing a gtf.

Quick question: is there no possibility of transferring selection from the rowStat table?

[kevinrue]

@plger

Quick question: is there no possibility of transferring selection from the rowStat table?

See https://github.com/csoneson/iSEE/issues/308#issuecomment-516546966, namely @LTLA reply to:

That said, I just checked and tables cannot be sources of selections.

Keep doing what you can in the current setup - it's good training - but know that we're preparing an "iSEE 2.0" which will change how custom panels are developed. We might even be able to throw in selection from tables, who knows. Everything in due time :)

[LTLA]

Transmission from tables to plots (or other tables) will be possible in 2.0.0. The identity of the transmitted points is determined by the search fields, either column-specific or global.

Technically speaking, it's not the same as a DT selection, which would involve actually clicking through the table to select multiple rows. None of the default Tables will enable multiple selections, mostly because the Table selections are linked to the other plots (e.g., the y-axis on FeatAssayPlots) and it's a pain to have to unselect your previous selection before making a new selection.

[kevinrue]

Whether search field or DT selection-based, it should be possible to add a couple of "Select all", "Unselect all" buttons that update the active rows. Anyway, something to investigate only once the main refactoring is completely done and documented.

[jorainer]

@plger , could you guide me to the part of the code you want me to check? Also, would be best if we continue detailed discussions then with issues in your repo.

[LTLA]

Whether search field or DT selection-based, it should be possible to add a couple of "Select all", "Unselect all" buttons that update the active rows.

Well, the default Table panels don't even have a concept of multiple manual selections. You would have to set selection="multiple" in datatable() for this to work, which runs into the usability problems that I mentioned above.