RFC010

Title: Normalizing Groups
Author: Lincoln Stein <lstein@cshl.org>
Version: 1
Date: 29 August 2001

It is not uncommon for genomic features to be grouped, and for the
groups themselves to be grouped into higher-ordered structures.
Familiar examples include: exons that are grouped into alternative
transcripts, transcripts that are grouped into genes, HSPs that are
grouped into gapped alignments.

The obvious way to handle these groupings is with a denormalized,
hierarchical schema like this one:

  <feature type="transcript" id="transcript1">
    <parts>
      <subfeature type="exon" id="exon1" />
      <subfeature type="exon" id="exon3" />
    </parts>
  </feature>

  <feature type="transcript" id="transcript2">
    <subfeature type="exon" id="exon1" />
    <subfeature type="exon" id="exon2" />
  </feature>

However, there's a problem with this, which is that the information
for exon1 is repeated twice.  If there's additional information
associated with exon1 (for example, external links), then the client
cannot know that it has collected all the information about exon1
until it has collected all the transcripts that contain it.

Another way to handle this is to normalize the groups:

  <feature type="exon" id="exon1" link="foo" />
  <feature type="exon" id="exon2" link="bar" />
  <feature type="exon" id="exon3" link="baz" />
  <group type="transcript" id="transcript1" link="foobar">
    <parts>
       <part type="exon" id="exon1" />
       <part type="exon" id="exon3" />
    </parts>
  </group>
  <group type="transcript" id="transcript2" link="bazzbuzz">
    <parts>
       <part type="exon" id="exon1" />
       <part type="exon" id="exon2" />
    </parts>
  </group>

In this way all the low-level features are declared explicitly and
given an ID that is unique (or at least unique for the scope of the
XML).  The low-level features are then explicitly grouped. Groups IDs
can be further grouped, for example:

  <group type="gene" id="geneA" link="barzz">
     <parts>
       <part type="transcript" id="transcript1">
       <part type="transcript" id="transcript2">
     </parts>
  </group>

The problem with this approach is that it requires the client to do
more bookeeping.  However in the long run I think it makes the system
more stable.