Hi James,
On 15•Jul•2009, at 3:12 PM, James Taylor wrote:
On Jul 15, 2009, at 8:43 AM, Pieter Neerincx wrote:
> The following is just a bunch of thoughts after using Galaxy for a
> while and which might be interesting for future developments...
Thanks for the suggestions, lots of good ideas here.
:)
> * There are three nice icons at the top of all my dataset items
in
> the
> history panel on the right for view, edit and delete. So why is there
> no save icon at the same location instead of a link further down?
This is simply because there is limited space. We've made the three
most frequently needed options available as icons, but there is not
space to include all options (save is just one, for certain
datatypes we also have browse links and other options). I've played
with a few ideas on how to make these options easier to get to. The
most likely candidate is adding a popup menu, but this needs further
study since any change to the history has potential to confuse
current users.
I guessed it was due to space constraints and apart from buying larger
screens I don't have a quick solution... I like the current 3 icons as
they give me quick access to functionality without expanding the
history item :), so I hope these will not be hidden in a popup menu
requiring extra clicking, but I like the idea of an extra icon for a
popup menu that gives me access to the less frequently used options.
Whatever you choose, anything that makes it more consistent should
make it easier to use :)
> * When I edit a workflow there is a save button above the canvas
and
> there is another on in the panel on the right when I edit the
> properties of a specific workflow item. As far as I can tell these
> buttons are not completely redundant, but why do I need two save
> buttons?
The reason is that one save button just saves and validates the
parameters for a single step, the other saves and validates
parameters for the entire workflow. We want to make sure that each
step is validated at the time it is being edited. Otherwise we would
need to guide the user through a complex validation project at save
time.
I'm not happy with the way his works right now either, but again
have not found a better solution. We could autosave the forms
whenever a user changes a field, but this has some tricky UI
implications.
I understand. How about naming the button that "saves" and validates
params for a single step "update" or "apply". Instead of saving the
workflow after each step, you could work with a copy of the workflow.
When changes are applied to a single step it updates the copy. Only
when you click save for the entire workflow the updated copy
overwrites the original. This way a user can also explore several
modifications and when the result is not satisfying cancel all mods by
simply not saving the workflow.
> Ensembl version 48, or UniProt 3 or some version of a reference
> assembly, etc. Or I might want to see how the results changed for a
> certain gene over time as result of updated databases and /or tools.
> So I might want to say to Galaxy show me all histories containing
> ENSGALG000012589 or NM_45689725. Hence, I'd love to be able to search
> histories.
This is a great idea, but computationally complex. If we just want
to search the metadata in the histories (title, info, etc) that is
pretty straightforward. Searching the actual data would be very
costly, since it would require some sort of full text indexing of
every dataset in Galaxy. Of course, this could be done in the
background rather than in realtime... worth considering.
I understand that searching the actual data could potentially be very
resource intensive, but I would already be very happy if I can search
the metadata. Simply being able to search for the analysis that was
based on a certain DB version would be really nice. Now I have to
lookup release dates for DB versions and based on that examine
histories a date near the DB release date.
What would make this work better is if we could have additional
metadata automatically added to datasets we get from external
datasources that would include the sort of information people would
be interested in searching for.
That would be great!
> In addition to make it a bit easier to trace thing in
> browse mode it would be nice if the date a history was last modified
> would be visible. Currently I only have the age of the history in
> minutes, hours or days. That is convenient for recent items, but for
> things that are longer ago a date makes more sense to me....
This is no problem, we can display a date if it is more than a few
days old.
I'm a bit confused: Does that mean you could add this to a future
Galaxy version or is it already possible with the current version and
did I fail to find some configuration parameter?
> * There is a fixed "Database/Build" popup that I can
use to tag my
> data sets, but this feels artificially limited. Is there any reason
> why the species and database version cannot be separate items? If
> there would be a popup first to select a species followed by a second
> popup to select the genome assembly version, the lists could be a lot
> smaller and hence easier to navigate.
Absolutely. We're playing with a variety of ways to make database/
build easier to work with. The main problem is that since we draw
builds from multiple sources, getting the appropriate data to group
them in a reliable way is challenging.
> In addition there are cases where I do have a species, but don't
> have
> an assembly or where there are additional version numbers to keep
> track of.
> For example I have lots of Ensembl data. Ensembl does not have a
> single version number, but 3 version numbers. There is one for the
> database schema, one for the assembly and one for the annotation/
> genebuild. The curent version for mouse is for example: 55 37 h,
> where
> 55 is the release and schema version number, 37 the assembly and "h"
> the version of the gene build. In addition I recently moved to a
> proteomics group and might want to capture DB version numbers for
> species without a reference assembly. for example I might know the
> species name and the fact I'm using UniProt 15.5... but currently I
> cannot easily capture that in a consistent way. (I know I might add
> this to the "info" for a dataset, but it's free text, with all kinds
> of possible spelling variants as a result...)
Builds and versioning are one of the trickiest problems we have to
deal with. I'd love to hear your suggestions on how you'd like the
tracking of this information to work.
I know this isn't easy... Preferably you would like to be able to do
things like search for analyses that were done with DB versions older
or newer than a certain version. If all data sources would simply use
integers and increment with 1 for each release, it would be easy. A
partial solution might be capturing release dates in a separate field
as you can easily standardise the format for a date field and search
it :). Would I would love to see is separate metadata field for
species, database, version number and release date. Of these fields at
least the species, database and release date fields could use a
thesaurus / standardised formatting making them easily searchable...
Cheers,
Pi
Thanks,
James
-------------------------------------------------------------
Biomolecular Mass Spectrometry and Proteomics
Utrecht University
Visiting address:
H.R. Kruyt building room O607
Padualaan 8
3584 CH Utrecht
The Netherlands
Mail address:
P.O. box 80.082
3508 TB Utrecht
The Netherlands
phone: +31 (0)6-143 66 783
email: pieter.neerincx(a)gmail.com
skype: pieter.online
------------------------------------------------------------