Zheng's notes on wormbase migration

Jump to: navigation, search


this page is a record of my experience for migrating wormbase onto chado. As far as I know, wormbase is based on the Acedb (an object-oriented schema) mapping onto rmdbs (mysql/postgresql). Chado is a new, more sophisticated, but generic schema. this page has been quite long and I am moving from gff3 bulkload to xmlxort, so I will just keep this one here for record but not update it. the migration will continue at learn XMLXORT

module-based migration

focus on sequence module first. using gff3 files as input.

gff3 columns and chado table relationship
gff3 chado table
seqid featureloc.srcfeature_id, featureloc.rank=0
source feature_dbxref.dbxref_id, dbxref.db_id(db.name='GFF source')/accession/version
type feature.type_id, cvterm.cvterm_id/dbxref_id, cv.cv_id, cvterm_dbxref.cvterm_id
start featureloc.fmin
end featureloc.fmax
strand featureloc.strand
phase featureloc.phase
attribute ID feature.name, feature.uniquename if ID is unique otherwise 'auto'+feature.feature_id
attribute Name feature.name
attribute Alias feature_synonym.synonym_id, synonym.name/synonym_sgml type_id(cvterm.cvterm_id for syn)
attribute Dbxref feature_dbxref, another dbxref, see column3(type)
attribute Gap
attribute Note featureprop
custom tag(lower case) db.name=null, cv.name=local, dbxref.accession='autocreated:xxx'
Ontology_term feature_cvterm.cvterm_id, feature_cvterm_dbxref.feature_cvterm_id, feature_cvterm_pub, feature_cvtermprop

feature.dbxref_id is nullable. Dbxref could be lower-case

bio-chaos and gmod_bulk_load_gff3

both bio-chaos 0.02 and gmod_bulk_load_gff3 can theoretically work. btw, bio-chaos 0.01 is included in the schema cvs download, but no gff3->chaos script in it. so go to bio-chaos 0.02 for prerequisite and installation. read a book XML in a nutshell helps a lot for me to understand chaos DTD.

Now I know XMLXORT will be finally used not only for sequence-related data but also for other data, I have to learn XMLXORT.

first step

get the current release WS171 gff3 file from wormbase. total 1.07G. split it by:

grep -P /^I\t/
[zha@localhost 1]$ ls -l chrI.gff3
-rw-rw-r-- 1 zha zha 165530115 Mar 20 17:33 chrI.gff3

only two directive lines in ws171

##gff-version 3
##Index-subfeature 0

but adding the size of chr-based files does not (similarly) equal to the original size of ws171, ??? I lost something here already?

pain for loading

  • first try load a sample gff3

a sample nGASP gff3 file has been successfully transformed to chadoXML by bio-chaos.

use Bio::Chaos;
my $path = '/home/zha/gff3/phase2_confirmed.gff3';
my $infmt = 'gff3';
my $outfmt = 'chadoxml';
my $c = Bio::Chaos->new;
$c->parse($path, $infmt);
print $c->transform_to($outfmt)->xml;

but I doubt it could load onto chado for the following test on gmod-bulk-load-gff3.

[zha@localhost gff3]$ gmod_bulk_load_gff3.pl --dbname zha --organism worm --gfffile  \ phase2_confirmed.gff3
Preparing data for inserting into the zha database
(This may take a while ...)
Unable to find srcfeature IV in the database.

sort it so that Parent of a feature (column 9 tag Parent) comes before the feature line in file. sorted it by:

gmod_sort_gff3 --infile chrI.gff3 > chrI.unresolved

two files are generated:


but adding the size of them, much less than the size of chrI.gff3, I definitely lost a lot here, abadon this is not what I expected from the name of the file and perldoc.

my experience with chromosome I

  • chromosome definition line
I       Link    chromosome      1       15072419        .       +       .       Name=I

I manually changed it to

I       Link    chromosome      1       15072419        .       +       .       ID=I, Name=I

and put it at the top of the gff3 file, it is NOT a problem of gff3 file, i.e., the file is valid wherever this line is or even without this line, but put it on top helps the bulk_load, or maybe gmod_gff3_prepocessor will try to do this change.

  • clone_end line
I       .       clone_end       10038617        10038617        .       .       .       Name=C03C11
no cvterm for clone_end at /usr/lib/perl5/site_perl/5.8.8/Bio/GMOD/DB/Adapter.pm line 3445, <GEN0>  line 12402.
Issuing rollback() for database handle being DESTROY'd without explicit disconnect().

this is a valid line, i.e, clone_end is a valid SOFA term, accroding to SOFA v2 (05-16-2005). what we loaded in chado installation is the SO latest minor revision version v2.1 (08-16-2006). in this version clone_end change to clone_insert_end.

  • this is a known situation...
Your GFF3 file uses a tag called 'confirmed_est', but this term is not
already in the cvterm table so that it's value can be inserted
into the featureprop table.  The easiest way to rectify this is
to execute the following SQL commands in the psql shell:
 INSERT INTO dbxref (db_id,accession)
   VALUES ((select db_id from db where name='null'),'autocreated:confirmed_est');
 INSERT INTO cvterm (cv_id,name,dbxref_id)
   VALUES ((select cv_id from cv where name='autocreated'), 'confirmed_est',
           (select dbxref_id from dbxref where accession='autocreated:confirmed_est'));
and then rerun this loader.  Your other option is to
write a special handler for this tag so that it will
go where you want it in the database.
Died at /usr/lib/perl5/site_perl/5.8.8/Bio/GMOD/DB/Adapter.pm line 2834, <GEN0> line 13204.
Issuing rollback() for database handle being DESTROY'd without explicit disconnect().

Noticed the above situation the cvterm is in column 3 (type), here the term is in column 9, a tag, such as ID, NAME, Dbxref, etc. I encoutered a series of them, which are good information.

non-canonical tag used in WS171 chrI
predicted ncrna gene

Scott suggests write a local ontology, such as wormbase ontology; Or as Don suggests, automatically load it. Notice this lower case column 9 tag may have some terms exactly the same as column 3 SOFA cvterm, but they are treated as different ones. I removed all lines like this one

I       ncRNA   ncRNA   10010373        10010484        999.545 -       .       Name=Note;predicted ncrna    gene=1;rnaz-512263=RNAz-512263:Note

also this one

I       Coding_transcript       mRNA    11877789        11887256        .
+       .       ID=Transcript:Y47H9C.5a.1;Name=Y47H9C.5a.1;Note=DnaJ domain%3BWo
rmPep WP:CE20265%3BNote dnj-27%3BPrediction_status Partially_confirmed%3BGene WB
Gene00001045%3BCDS Y47H9C.5a%3BThioredoxin;1=1
  • several options

--analysis for file that has analysis (eg. blast) feature; --noexon, this is a 'weird' option. Since chado treat CDS as intersection between exons and transcript polypeptides, CDS and UTR lines are converted to generate exon and polypeptide by default in bulk_load. if gff file has exon lines, then you shall use this option, which means do not generate exon (?, since you have exon lines, but the option is named noexon, tricky). my command becomes:

gmod_bulk_load_gff3.pl --dbname zha --organism worm --analysis --gfffile chrI_2.gff3.sorted --noexon
  • non-central-dogma-feature

this is what I think: shall have at least 3 lines for compliance with central dogma, a gene line, a mrna line and a cds line. the following warning is actually casued by lacking a gene line. this is what I observed:

There is a CDS feature with no parent (ID:)  I think that is wrong!
This GFF file has CDS and/or UTR features that do not belong to a
'central dogma' gene (ie, gene/transcript/CDS).  The features of
this type are being stored in the database as is.
no parent CDS:B0019.1:wp90;

since the file has this two lines:

I       history mRNA    12759743        12764935        .       -       2       ID=CDS:B0019.1:wp90;Name=B0019.1:wp90;Indexed=1
I       history CDS     12764810        12764935        .       -       0       Parent=CDS:B0019.1:wp90

although the warning says 'CDS feature', but it could also be lacking a gene line that trigger the warning. Infact, all the lines with 'history' in column 2 do not have a corresponding line with 'gene' in column 3.