On 12/03/09 3:46 PM, "David Riley" <[email protected]> wrote:
> Hi all,
>
> I am very new to BioMart and my goal is to use it to build marts on a Chado
> instance for the Sybil package (sybil.sf.net <http://sybil.sf.net> ). My goal
> is to build a fairly comprehensive and easy to use search builder using extjs
> with BioMart driving the search data sets and filters. I'm just starting out
> but I have already run into a bit of a snag.
>
> I am trying to connect Martbuilder 0.7 to a Postgres 8.1 Chado instance. All
> the tables seem to get pulled down properly but non of the foreign key
> relationships or unique key constraints. The keyguessing mechanism works
> alright except there are several instances where a table has a type_id column
> that needs to be linked to a cvterm_id column and these are lost and kind of
> integral to the schema.
>
> I am wondering if anyone is currently using BioMart with a Chado instance and
> if so how.
>
> I have also noticed that even doing the 'give me the first 10 rows' causes an
> OutOfMemory exception. The database I am trying to connect to has 7,607,708
> rows in feature and I am wondering if this is a bit bigger than BioMart was
> designed for.
>
> I noticed some warnings about Java 1.6 which happens to be the version I am
> using so this may explain the memory problem.
>
> Any help and/or Chado BioMart experiences would be greatly appreciated.
>
> Thanks,
>
> Dave
Hi Dave,
A few comments here:
1. Building relations from postgres does not depend on keyguessing (this is
only for mysql myisam). Postgres PK/FK gets resolved through database
metadata so in theory nothing should get lost. If you could point us to
where we could get this schema we can double check it here.
2. The problem with out of memory is likely due to DMD implementation of
'limit to x rows' on which Mbuilder is dependent on. As far as I remember
this implementation takes all the results first and then let you see first
10 which in this case would not work well ;) but I don't want to guess here.
We are going to check that and let you know what's going on there.
3. Mbuilder was not officially tested on 1.6 so yes you may see some warning
there.
4. Finally Mbuilder is not yet clever enough to do a meaningful
transformation of chado into mart. (it will work but what you get out of
this may not be as useful for you as you think). This is because the current
algorithm only works well with schemas which are strictly 3NF. We are
planning to implement a new flavour of the chado like transformation soon.
Anyway we'll check all the problems and will get back to you soon
A.
