<!--    The Gene Ontology DTD                                   -->

<!--  The idea is that GO is dumped as a linear list of terms.  -->
<!--  The relationships between the terms are represented in    -->
<!--  RDF, using the rdf:about and rdf:resource tags.  RDF was  -->
<!--  chosen because of its flexibilty at representing graphs   --> 
<!--  and widespread tool support.                              -->

<!--  Basically, one can think of rdf:about and rdf:resource    -->
<!--  as being similar to xml's id and idref, respectively.     -->
<!--  RDF, however, can represent multiple parentage.  The      -->
<!--  other thing to know about rdf is that the IDs should be   -->
<!--  universally unique, so GO:0003700 becomes:                -->
<!--  http://www.geneontology.org/go#GO:0003700                 -->

<!ENTITY parseType "CDATA  #FIXED  'Resource'">

<!ELEMENT go:go (
    go:version?,
    rdf:RDF
)>

<!ATTLIST go:go
    xmlns:go   CDATA  #FIXED  "http://www.geneontology.org/dtd/go.dtd#"
    xmlns:rdf  CDATA  #FIXED  "http://www.w3.org/1999/02/22-rdf-syntax-ns#"
>

<!ELEMENT go:version (#PCDATA)>

<!ATTLIST go:version
    timestamp  CDATA  #IMPLIED
>


<!--  RDF requires any rdf nodes to be inside the rdf:RDF tag   -->

<!ELEMENT rdf:RDF (go:term*)>
    
<!ELEMENT go:term (
    go:accession,
    go:name,
    go:synonym*, 
    go:definition?, 
    go:comment*,
    (go:part_of | go:is_a | go:negatively_regulates | go:positively_regulates | go:regulates)*,
    go:dbxref*,
    go:association*,
    go:negative_association*,
    go:history*,
    go:is_obsolete*
    )>

<!ATTLIST go:term
    focus   (yes|no)  #IMPLIED
    rdf:about  CDATA     #REQUIRED
    >

<!ELEMENT go:accession (#PCDATA)>
<!ELEMENT go:name (#PCDATA)>
<!ELEMENT go:is_obsolete (#PCDATA)>
<!ELEMENT go:comment (#PCDATA)>

<!--  Synonyms are simply represented as a text string.   -->

<!ELEMENT go:synonym (#PCDATA)>
<!ELEMENT go:definition (#PCDATA)>

<!--  Each term can be is_a and part_of many terms  -->
<!--  This is represented with an rdf:resource att -->
<!--  Right now, you can also add a common name    -->
<!--  for the linked term.  Is this smart?         -->

<!ELEMENT go:part_of (#PCDATA)>
<!ATTLIST go:part_of
  rdf:resource   CDATA   #REQUIRED
>

<!ELEMENT go:regulates (#PCDATA)>
<!ATTLIST go:regulates
  rdf:resource   CDATA   #REQUIRED
>

<!ELEMENT go:negatively_regulates (#PCDATA)>
<!ATTLIST go:negatively_regulates
  rdf:resource   CDATA   #REQUIRED
>

<!ELEMENT go:positively_regulates (#PCDATA)>
<!ATTLIST go:positively_regulates
  rdf:resource   CDATA   #REQUIRED
>

<!ELEMENT go:is_a (#PCDATA)>
<!ATTLIST go:is_a
  rdf:resource   CDATA   #REQUIRED
>

<!ELEMENT go:association (
   go:evidence+,
   go:gene_product,
   go:association_qualifier*
)>

<!ELEMENT go:negative_association (
   go:evidence+,
   go:gene_product,
   go:association_qualifier*
)>

<!ATTLIST go:association
   rdf:parseType CDATA  #FIXED  'Resource'
>

<!ELEMENT go:association_qualifier (
  go:qualifier_name+
)>

<!ELEMENT go:qualifier_name (#PCDATA)>

<!ELEMENT go:evidence (
    go:dbxref?
)>


<!--  IC, ND and NR really do not belong, but the curators  -->
<!-- are using them, so I'm including them for now          -->
<!-- for the sake of completeness.                          -->

<!ATTLIST go:evidence 
    evidence_code   ( IEA | IMP | IGI | IPI | ISS | IDA | IEP | TAS | NAS | IC | ND | NR | RCA | EXP | IBA | IGC | IKR | IMR | IRD | ISA | ISM | ISO | NULL )  #REQUIRED
>


<!ELEMENT go:gene_product (
    go:name,
    go:dbxref
)>

<!ATTLIST go:gene_product
    rdf:parseType CDATA  #FIXED  'Resource'
>

<!--  I'm using reference instead of accession  -->
<!--  since these don't have to be primary db   -->
<!--  accessions in the traditional sense.      -->
<!--  Do they?                                  -->

<!ELEMENT go:dbxref (
    go:database_symbol,
    go:reference
)>

<!ATTLIST go:dbxref
    rdf:parseType CDATA  #FIXED  'Resource'
>

<!ELEMENT go:database_symbol (#PCDATA)>

<!ELEMENT go:reference (#PCDATA)>
<!ATTLIST go:reference
    type CDATA  #IMPLIED
>

<!--  Each term should have one history node at   -->
<!--  creation.  Another history node is added    -->
<!--  each time a term is deleted, split, moved   -->
<!--  or merged.                                  -->

<!ELEMENT go:history (
    go:date, 
    go:person, 
    go:rationale, 
    ( go:created | 
      go:deleted | 
      go:split   | 
      go:merged  | 
      go:moved )
    )>
<!ATTLIST go:history
  obsolete   (yes|no)   #IMPLIED
>


<!ELEMENT go:date (#PCDATA)>
<!ELEMENT go:person (#PCDATA)>
<!ELEMENT go:rationale (#PCDATA)>
<!ELEMENT go:created (EMPTY)>
<!ELEMENT go:deleted (EMPTY)>

<!--  In go:split, go:merged and go:moved, the  -->
<!--  rdf:resource ref should point to the      -->
<!--  "parent" term, ie the term that this      -->
<!--  term was split from.                           -->

<!ELEMENT go:split (EMPTY)>
<!ATTLIST go:split
  rdf:resource   CDATA   #REQUIRED
>
           
<!ELEMENT go:merged (EMPTY)>
<!ATTLIST go:merged
  rdf:resource    CDATA   #REQUIRED
>
              
<!ELEMENT go:moved (EMPTY)>
<!ATTLIST go:moved
  rdf:resource     CDATA   #REQUIRED
>