Gene Ontology Meeting February 25-26, 2000 at Astra-Zeneca in Cambridge, MA

Attendees:

Michael Ashburner (FlyBase)
Suzanna Lewis  (FlyBase)
Heather Butler (FlyBase)

Judy Blake (MGI)
Janan Eppig (MGI)
David Hill (MGI)
Joel Richardson (MGI)
Martin Ringwald (MGI)
Allan Peter Davis (MGI)

Michael Rebhan (Astra Zeneca)

Mike Cherry (SGD)
Cathy Ball (SGD)
Midori Harris (SGD)
Andrew Kasarskis (SGD)

AGENDA ITEMS

Progress Reports
Celera Report
Papers
Collaborators and Other Projects
Ontology Issues
Style and Work Practices
Tools for GO
Questions from Michael R.
Plans for Next Meeting

PROGRESS REPORTS

Mouse folks:

Judy Blake has submitted the GO grant.

The mouse members have assigned approximately 4500 genes to GO terms.

100 by hand
650 by EC number
1270 using Swiss-Prot
2500 using mouse nomenclature

Without counting the Swiss-Prot data, they used 474 Molecular Function
terms, 50 Cellular Component terms and 80 Biological Process terms.

Since they are using automated annotation, they have performed a
variety of quality checks, such as looking for more than one
annotation within an ontology. They have come close to exhausting the
current automated assignments and are going to be doing more by hand
in the future.

Yeast Folks:

SGD has 1524 genes in the gene association file.  About 1000 of these
are ORFs and the rest are tRNAs or snoRNAs.  All SGD genes have been
GO-annotated by hand.

Fly Folks:

FlyBase currently has about 3000 genes annotated mostly by hand.
Heather has worked through the protein kinases and will next tackle
the protein phosphatases.  Annotation of new genes will be largely
done by sequence similarity, while existing genes will be done by hand
in related chunks.  When the Drosophila sequence is released in March,
there will be a large amount of sequences annotated to a high-level GO
ID.  These will be deepened to more specific GO nodes with time.

CELERA REPORT

GO was used in the annotation of the Drosophila genome at Celera.
Suzanna made a dataset with all genes annotated to the molecular
function GO and used it for BLAST searches.  Usually, the level of GO
node was quite high -- only one or two terms from the top.  Where
experts in a field were expected to be annotating genes, the
specificity of the GO nodes used were increased (for example,
olfactory receptors).  Ultimately, there were 40 bins labelled by GO
name (the 40th was "unknown").  Annotators were then able to have a
pretty reasonable guess as to the function of the new fly gene.  A
second binning with biological process and cellular component showed a
terrific correlation with the first.

About half the genes from the Celera set are associated with a GO
term.  Since a given gene has a less than 50% chance of having been
seen by a human, an association with a GO term is very valuable.

FlyBase is still waiting to receive the sequence -- it will be
released with the publication of the papers in March.  FlyBase will be
responsible for updating the sequence in GenBank.

PAPERS

We agreed to immediately pursue three publications:

1) Nature Genetics solicited a short (2500 word) article from David
Botstein.  It will be submitted March 10, with a short author list.

2) Genome Biology -- Michael Ashburner has been asked to write a short
(1000 word) article for their premier issue.  It will most likely have
an authorship along the lines of "The GO Consortium".

3) Genome Research -- Judy Blake will adapt the grant to a "big" paper
to be submitted to Genome Research.

4) NAR database issue -- We will submit a paper to NAR as a matter of
course.  The submission won't be until August or September.  Since
there are likely to be changes in the NAR policies, we will discuss
the details of the NAR paper at the next meeting.

COLLABORATORS AND OTHER PROJECTS

There was a great deal of discussion about taking on other organisms
and collaborators.  The conclusion was that before we take on other
organisms, we must first meet the following goals:

--We need to be in a database (Suzanna Lewis will be working on this,
with help from Joel Richardson).  Hopefully, this will be accomplished
by the next meeting.  See "Plans for Next Meeting" for more detailed
steps.

--Documentation of philosophy, styles and practices needs to be
written to record and communicate our current thinking.  See "Plans
for Next Meeting" for more detailed steps.

--A "GO manager" to coordinate changes to the ontology, arrange
training, communicate with all groups, etc needs to be hired.  Midori
Harris has volunteered to assume the responsibility.

Michael Ashburner suggested we have two classes of partners -- the
first with write permission and the second without it.  These "second
class" partners will have to funnel suggestions and comments through a
full partner.

Other organism groups that have expressed interest include worm,
Arabidopsis, and S. pombe.  We'll invite a representative from the
worm and Arabidopsis database groups to the next GO meeting.

Michael Ashburner has received a grant application for "BioBabel" -- a
proposal to adopt GO terms within SwissProt, Enzyme Commission and
Interpro.  Representatives from this group can also be invited to the
next meeting.

ONTOLOGY ISSUES

Methods and practices for editing and maintaining the ontology took up
a large portion of the discussions.  Conclusions will be listed, and
in the cases where the discussion is particularly illuminating, the
discarded options will be listed as well.

1) Changes to GO nodes that have multiple parents... When editing one
of the ontologies, it is more convenient to add another node in only
one position.  For example, if we start with the structure shown
below:

a
  b
    d
      e
      f

If we want to add node 'c' as a of 'd' and a child of node 'a', do we
need to edit all the appropriate lines, or just one?  The group
decided to make an "editable" non-redundant version of the ontologies:

Linear, redundant format (for viewing):

a
  b
    d
      e
      f
  c
    d
      e
      f

Non-redundant format (for editing):

a
  b
    d  % c
      e
      f

The envisioned procedure is that a curator checks out the compressed,
or non-redundant, version and then views an expanded version using a
planned tool we're calling "The Validator."  When an edit needs to be
made to an ontology, it is made in the compressed version and tested
with the Validator.  The compressed version is then checked back into
the cvs.  The Validator will be written by Joel, suing specifications
mentioned later.

The web will display the expanded, read-only format.

2) We will add GO id to parent terms.  For example, we used to state:

term1 ; GOID1 % term2

Now we will state:

term1 ; GOID1 % term2 ; GOID2

3) GO nodes should aggressively avoid using species-specific
definitions.  We agreed to substitute "Yeast mating" with "Mating,
sensu Saccharomyces."  Using the "sensu" reference makes the node
available to other species that use the same
process/function/component.  Each organism database will take care of
their contributions to the species-specific language.

4) We will get rid of cellular component references in the function
ontology.  For example, "mitochondrial primase" needs only be
"primase."  There are many cases where component terms are appropriate
in the process ontology, so those will remain.  Michael A. will take
care of this.


5) Joel pointed out these logical relationships that we need to make
sure are true in the ontologies:

if A is part of B
and C isa B,
is A part of C? --- YES

if A is a B
and B isa C,
is A isa C? --- YES

if A is part of B
and B is part of C,
is A part of C? --- YES

if A isa B
and C is part of B,
is C part of A?  --- NOT NECESSARILY

Joel will send out a list of the logical inconsistencies that he has
detected.

6) An example that got a lot of attention is the case of the mitotic
chromosome's location in the cellular component ontology.  While the
mitotic chromosome resides in the nucleus in yeast, it is cytoplasmic
at this stage of cell life in mouse or fly.  In addition, many
organisms have chromosomes that are NOT located in the nucleus.  The
solution arrived at was to remove chromosome from the nucleus in
general and place the appropriate subsets of chromosomes in the
correct place (nuclear, cytoplasmic, mitochondrial).

7) We need to track deleted GO ids.  There are types of things that
can happen to GO terms -- merging two (or more) nodes, splitting a
node, deleting a term.

a. When a term is deleted, we will cut the line out and paste it at
the end of the file (or as a child of the parent "defunct", I don't
recall the final decision), using the following format:

<defunct mitochondrial primase ; GO:idnumber

b. When a term is split, giving two new GO ids, use the following format:

<defunct split term, see GO:idnumber1 and GO:idnumber2 ; GO:idnumber

c. To merge two nodes, there is no need to delete one.  The format:

<term A ; GO:idnumbera ; GO:idnumberb ; synonym:termb

8) Parts of the molecular function ontology can be given
reality-checks with Monica Riley's enzyme hierarchy and Milton Saier's
transporter hierarchy.

9) One of the results of the discussion of Michael A. and Andrew's
style document was an agreement that we should strive to standardize
the ways words are used and concepts are subdivided within the
ontologies.  Joel brought up the possibility of separating branches of
an ontology.  His example "deconstructed" carbohydrate metabolism into
two DAGs, one for metabolism and one for carbohydrates.


STYLE AND WORK PRACTICES

There was good general agreement that we need to consolidate and
document our styles and work habits, especially when we consider that
other groups are going to be coming aboard the project.

Andrew and Michael A. have been working on a style guide.  All the
suggestions were accepted.  Discussion led to more conclusions:

1) Unless you can come up with a GOOD definition to describe a node,
don't add a new node.  We also need to work hard at defining existing
nodes.

2) Something gets a GO node if more than one organism is involved, or
more than one protein is involved.

3) Sometimes it helps to think of a molecular function as something
that can be assayed in vitro, and a biological process as something
that can lead to a phenotype when disrupted.

4) EC numbers belong in the molecular function ontology, but not in
the cellular component ontology.

5) Cellular structures are not really functions.  For example, a
ribosomal protein is not a function, but being a structural component
of the ribosome is a function.

6) Watch carefully to make sure ALL appropriate GO IDs are assigned.
For example, if your function is protein kinase, then it's quite
likely that the process will include protein phosphorylation.

7) Watch for gene products that are not assigned to all three
ontologies.

8) SGD will add GO: to the gene associations file.

9) Curators can add sequence IDs to the justification of gene
associations (for example, GB:accessionnumber or PIR:accessionnumber).

10) In xml, we will superscript and subscript with the <sup></sup> and
<sub></sub> tags.

11) We currently cannot standardize rules for subdividing ontology
terms, but instead will continue to make each decision on a
case-by-case basis.

12) Gene products in themselves are not nodes of the function
ontology, although doing something with or to a specific gene product
can be one.  For example, being hedgehog is not likely to be a
function, but being a hedgehog receptor or hedgehog receptor ligand
are functions.

13) We may eventually need a synonym table to facilitate queries.

14) Changes that need to made to the ontology to meet the current
style include eliminating unnecessary hyphens, adjust grammar so that
"transporters" become described as "transport" and "transporting,"
remove words like "protein" and "factor" where we can be more
explicit.

15) Heather and Midori will write some documentation about the
evidence codes.

16) We need to think about a "best practices" document that will state
and explain good work habits for both current and future annotators.
In the meantime, we will share any help documents, such as SGD's
"Instructions for Annotating Genes Using GO."



TOOLS FOR GO

1) Database

Suzanna will get a handle on this.  The major difficulty has been
hiring a programmer.  Michael R. offered some help on this from
Astra-Zeneca.

Suzanna is planning on using MySQL to create a version to distribute
from the central site.  The schema is not yet ready, but Suzanna and
Joel will work on this together.  The database will also need the
ability for bulk load.

2) Validator - Joel will do this

We need a validator to check for:

a. cycles
b. deletion of nodes used in gene association files
c. syntactic correctness (refer to logical relationships described in the 
ONTOLOGY ISSUES section.)
d. unique IDs
e. warning message of the number of affected nodes
f. orphans
g. new nodes have IDs

Associated with the validator is the ability to compact and expand the
ontologies for writing and reading.  The validator will run on the
central site, as well as locally for checking before an edited
ontology is checked back in.  Joel plans on writing this in python, so
each site will need to install it.

3) GO BLAST server - Mike C. will take care of this


The GO BLAST server will use a dataset of GO-annotated protein
sequences.  The results should show each GO node associated with a
gene product, as well as a few generations of ancestors.

4) Annotation aids

It would be nice for curators to have a tool that, given a single
node, display all other gene products at that node (and nearby nodes)
as well as all their other GO associations.  This would assist
curators in assigning a gene product to as many GO terms as needed, by
showing them all other GO terms that might be related.

5) Suzanna's browser needs to be installed at Stanford, so we can all
be using it from the same server.

6) Michael R. suggested we make a link to a dtd (datatype definition)
file.  Suzanna will look into finding a tool that will read the xml
and create a dtd file.

ANSWERS TO QUESTIONS FROM MICHAEL R.

1) GO ids will be stable.  They may be "defuncted", but they will not
go away.

2) "is a" and "part of" are likely to be used for quite some time.
However, "part of" means "can be a part of", NOT "is always a part
of."

3) Incyte still expresses interest, but that's all we've received from
them.

4) Homepage recommendations -- Mike C. will add a bit from the grant
to add more detail to the homepage.  It might also benefit from the
addition of statistics from the gene association files.

5) Should we have an ftp site that allows one to download the most
recent version of GO?

6) Michael R. will create a FAQ to be linked from the home page.

7) Mike C. will put SGD's PowerPoint GO presentations to the GO site.

PLANS FOR THE NEXT MEETING

The next GO meeting will be in Cambridge, UK June 29 and 30.

The plans are:

1) Have documentation ready

a. GO philosophy document (Michael A., Judy and Midori)

b. Rules for making changes to GO (Michael A. and Andrew)

c. Rules for applying GO terms -- this is currently
project-specific. Each project needs to think about this and bring
something to the table next time.  This should also include
particularly illuminating examples, such as chitin synthesis, mitotic
chromosomes.  It should also emphasize how to avoid making GO nodes
too species-specific, and mention the logical aspects of inserting or
moving nodes.


2) Invite representatives from BioBabel, Arabidopsis, and C. elegans.

3) Have database in place

4) Create programs described above

5) Work HARD on adding more GO definitions.  We have permission to use
the Oxford Dictionary of Biochemistry and Molecular Biology.

6) Make the ontology edits mentioned above

7) Write three (!) papers