X-Git-Url: http://source.jalview.org/gitweb/?a=blobdiff_plain;ds=sidebyside;f=src%2Fjalview%2Fio%2FStockholmFile.java;h=34ce3bf99db93a2bdc75743c6ef4c5da66361096;hb=caf13e6a9dc82023cd8648420201e5ad60ce0393;hp=10ddf53cbc5eb5af5252b28ed31d147a17336a10;hpb=7bc226b58110fa26d9dbd3f0c78095d06909ffc3;p=jalview.git
diff --git a/src/jalview/io/StockholmFile.java b/src/jalview/io/StockholmFile.java
index 10ddf53..34ce3bf 100644
--- a/src/jalview/io/StockholmFile.java
+++ b/src/jalview/io/StockholmFile.java
@@ -1,20 +1,19 @@
/*
- * Jalview - A Sequence Alignment Editor and Viewer
- * Copyright (C) 2007 AM Waterhouse, J Procter, G Barton, M Clamp, S Searle
- *
- * This program is free software; you can redistribute it and/or
- * modify it under the terms of the GNU General Public License
- * as published by the Free Software Foundation; either version 2
- * of the License, or (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
+ * Jalview - A Sequence Alignment Editor and Viewer (Version 2.6)
+ * Copyright (C) 2010 J Procter, AM Waterhouse, G Barton, M Clamp, S Searle
+ *
+ * This file is part of Jalview.
+ *
+ * Jalview is free software: you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
+ *
+ * Jalview is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR
+ * PURPOSE. See the GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along with Jalview. If not, see .
*/
/*
* This extension was written by Benjamin Schuster-Boeckler at sanger.ac.uk
@@ -27,69 +26,84 @@ import java.util.*;
import com.stevesoft.pat.*;
import jalview.datamodel.*;
-//import org.apache.log4j.*;
+// import org.apache.log4j.*;
/**
- * This class is supposed to parse a Stockholm format file into Jalview
+ * This class is supposed to parse a Stockholm format file into Jalview There
+ * are TODOs in this class: we do not know what the database source and version
+ * is for the file when parsing the #GS= AC tag which associates accessions with
+ * sequences. Database references are also not parsed correctly: a separate
+ * reference string parser must be added to parse the database reference form
+ * into Jalview's local representation.
+ *
* @author bsb at sanger.ac.uk
- * @version 0.3
+ * @version 0.3 + jalview mods
+ *
*/
-public class StockholmFile
- extends AlignFile
+public class StockholmFile extends AlignFile
{
- //static Logger logger = Logger.getLogger("jalview.io.StockholmFile");
+ // static Logger logger = Logger.getLogger("jalview.io.StockholmFile");
public StockholmFile()
{
}
- public StockholmFile(String inFile, String type)
- throws IOException
+ public StockholmFile(String inFile, String type) throws IOException
{
super(inFile, type);
}
+ public StockholmFile(FileParse source) throws IOException
+ {
+ super(source);
+ }
+
public void initData()
{
super.initData();
}
/**
- * Parse a file in Stockholm format into Jalview's data model. The file has
- * to be passed at construction time
- * @throws IOException If there is an error with the input file
+ * Parse a file in Stockholm format into Jalview's data model. The file has to
+ * be passed at construction time
+ *
+ * @throws IOException
+ * If there is an error with the input file
*/
- public void parse()
- throws IOException
+ public void parse() throws IOException
{
+ StringBuffer treeString = new StringBuffer();
+ String treeName = null;
// --------------- Variable Definitions -------------------
String line;
String version;
- // String id;
- Hashtable alAnn = new Hashtable(); // Alignment wide annotations
+ // String id;
Hashtable seqAnn = new Hashtable(); // Sequence related annotations
Hashtable seqs = new Hashtable();
Regex p, r, rend, s, x;
// ------------------ Parsing File ----------------------
- // First, we have to check that this file has STOCKHOLM format, i.e. the first line must match
+ // First, we have to check that this file has STOCKHOLM format, i.e. the
+ // first line must match
r = new Regex("# STOCKHOLM ([\\d\\.]+)");
if (!r.search(nextLine()))
{
- throw new IOException("This file is not in valid STOCKHOLM format: First line does not contain '# STOCKHOLM'");
+ throw new IOException(
+ "This file is not in valid STOCKHOLM format: First line does not contain '# STOCKHOLM'");
}
else
{
version = r.stringMatched(1);
- //logger.debug("Stockholm version: " + version);
+ // logger.debug("Stockholm version: " + version);
}
-// We define some Regexes here that will be used regularily later
- rend = new Regex("\\/\\/"); // Find the end of an alignment
- p = new Regex("(\\S+)\\/(\\d+)\\-(\\d+)"); // split sequence id in id/from/to
- s = new Regex("(\\S+)\\s+(\\w{2})\\s+(.*)"); // Parses annotation subtype
+ // We define some Regexes here that will be used regularily later
+ rend = new Regex("^\\s*\\/\\/"); // Find the end of an alignment
+ p = new Regex("(\\S+)\\/(\\d+)\\-(\\d+)"); // split sequence id in
+ // id/from/to
+ s = new Regex("(\\S+)\\s+(\\S*)\\s+(.*)"); // Parses annotation subtype
r = new Regex("#=(G[FSRC]?)\\s+(.*)"); // Finds any annotation line
- x = new Regex("(\\S+)\\s+(\\S+)"); //split id from sequence
+ x = new Regex("(\\S+)\\s+(\\S+)"); // split id from sequence
rend.optimize();
p.optimize();
@@ -97,7 +111,7 @@ public class StockholmFile
r.optimize();
x.optimize();
- while ( (line = nextLine()) != null)
+ while ((line = nextLine()) != null)
{
if (line.length() == 0)
{
@@ -105,16 +119,16 @@ public class StockholmFile
}
if (rend.search(line))
{
-// End of the alignment, pass stuff back
+ // End of the alignment, pass stuff back
this.noSeqs = seqs.size();
- //logger.debug("Number of sequences: " + this.noSeqs);
+ // logger.debug("Number of sequences: " + this.noSeqs);
Enumeration accs = seqs.keys();
while (accs.hasMoreElements())
{
String acc = (String) accs.nextElement();
- //logger.debug("Processing sequence " + acc);
- String seq = (String) seqs.get(acc);
+ // logger.debug("Processing sequence " + acc);
+ String seq = (String) seqs.remove(acc);
if (maxLength < seq.length())
{
maxLength = seq.length();
@@ -122,6 +136,14 @@ public class StockholmFile
int start = 1;
int end = -1;
String sid = acc;
+ // Retrieve hash of annotations for this accession
+ Hashtable accAnnotations = null;
+
+ if (seqAnn != null && seqAnn.containsKey(acc))
+ {
+ accAnnotations = (Hashtable) seqAnn.remove(acc);
+ }
+
// Split accession in id and from/to
if (p.search(acc))
{
@@ -129,30 +151,51 @@ public class StockholmFile
start = Integer.parseInt(p.stringMatched(2));
end = Integer.parseInt(p.stringMatched(3));
}
- //logger.debug(sid + ", " + start + ", " + end);
+ // logger.debug(sid + ", " + start + ", " + end);
Sequence seqO = new Sequence(sid, seq, start, end);
+ // Add Description (if any)
+ if (accAnnotations != null && accAnnotations.containsKey("DE"))
+ {
+ String desc = (String) accAnnotations.get("DE");
+ seqO.setDescription((desc == null) ? "" : desc);
+ }
+ // Add DB References (if any)
+ if (accAnnotations != null && accAnnotations.containsKey("DR"))
+ {
+ String dbr = (String) accAnnotations.get("DR");
+ if (dbr != null && dbr.indexOf(";") > -1)
+ {
+ String src = dbr.substring(0, dbr.indexOf(";"));
+ String acn = dbr.substring(dbr.indexOf(";") + 1);
+ jalview.util.DBRefUtils.parseToDbRef(seqO, src, "0", acn);
+ // seqO.addDBRef(dbref);
+ }
+ }
Hashtable features = null;
// We need to adjust the positions of all features to account for gaps
try
{
- features = (Hashtable) ( (Hashtable) seqAnn.get(acc)).get(
- "features");
- }
- catch (java.lang.NullPointerException e)
+ features = (Hashtable) accAnnotations.remove("features");
+ } catch (java.lang.NullPointerException e)
{
- //loggerwarn("Getting Features for " + acc + ": " + e.getMessage());
- //continue;
+ // loggerwarn("Getting Features for " + acc + ": " +
+ // e.getMessage());
+ // continue;
}
// if we have features
if (features != null)
{
+ int posmap[] = seqO.findPositionMap();
Enumeration i = features.keys();
while (i.hasMoreElements())
{
+ // TODO: parse out secondary structure annotation as annotation
+ // row
+ // TODO: parse out scores as annotation row
+ // TODO: map coding region to core jalview feature types
String type = i.nextElement().toString();
- Hashtable content = (Hashtable) features.get(type);
-
+ Hashtable content = (Hashtable) features.remove(type);
Enumeration j = content.keys();
while (j.hasMoreElements())
{
@@ -162,13 +205,17 @@ public class StockholmFile
for (int k = 0; k < byChar.length; k++)
{
char c = byChar[k];
- if (! (c == ' ' || c == '_' ||
- c == '-'))
+ if (!(c == ' ' || c == '_' || c == '-' || c == '.')) // PFAM
+ // uses
+ // '.'
+ // for
+ // feature
+ // background
{
- int new_pos = seqO.findPosition(k);
- SequenceFeature feat =
- new SequenceFeature(type,
- desc, new_pos, new_pos, 0f, null);
+ int new_pos = posmap[k]; // look up nearest seqeunce
+ // position to this column
+ SequenceFeature feat = new SequenceFeature(type, desc,
+ new_pos, new_pos, 0f, null);
seqO.addSequenceFeature(feat);
}
@@ -178,18 +225,22 @@ public class StockholmFile
}
}
- //logger.debug("Adding seq " + acc + " from " + start + " to " + end + ": " + seq);
+ // garbage collect
+
+ // logger.debug("Adding seq " + acc + " from " + start + " to " + end
+ // + ": " + seq);
this.seqs.addElement(seqO);
}
+ return; // finished parsing this segment of source
}
else if (!r.search(line))
{
- //System.err.println("Found sequence line: " + line);
+ // System.err.println("Found sequence line: " + line);
// Split sequence in sequence and accession parts
if (!x.search(line))
{
- //logger.error("Could not parse sequence line: " + line);
+ // logger.error("Could not parse sequence line: " + line);
throw new IOException("Could not parse sequence line: " + line);
}
String ns = (String) seqs.get(x.stringMatched(1));
@@ -206,80 +257,88 @@ public class StockholmFile
String annType = r.stringMatched(1);
String annContent = r.stringMatched(2);
- //System.err.println("type:" + annType + " content: " + annContent);
+ // System.err.println("type:" + annType + " content: " + annContent);
if (annType.equals("GF"))
{
- /* Generic per-File annotation, free text
- * Magic features:
- * #=GF NH
- * #=GF TN
- * Pfam descriptions:
- 7. DESCRIPTION OF FIELDS
-
- Compulsory fields:
- ------------------
-
- AC Accession number: Accession number in form PFxxxxx.version or PBxxxxxx.
- ID Identification: One word name for family.
- DE Definition: Short description of family.
- AU Author: Authors of the entry.
- SE Source of seed: The source suggesting the seed members belong to one family.
- GA Gathering method: Search threshold to build the full alignment.
- TC Trusted Cutoff: Lowest sequence score and domain score of match in the full alignment.
- NC Noise Cutoff: Highest sequence score and domain score of match not in full alignment.
- TP Type: Type of family -- presently Family, Domain, Motif or Repeat.
- SQ Sequence: Number of sequences in alignment.
- AM Alignment Method The order ls and fs hits are aligned to the model to build the full align.
- // End of alignment.
-
- Optional fields:
- ----------------
-
- DC Database Comment: Comment about database reference.
- DR Database Reference: Reference to external database.
- RC Reference Comment: Comment about literature reference.
- RN Reference Number: Reference Number.
- RM Reference Medline: Eight digit medline UI number.
- RT Reference Title: Reference Title.
- RA Reference Author: Reference Author
- RL Reference Location: Journal location.
- PI Previous identifier: Record of all previous ID lines.
- KW Keywords: Keywords.
- CC Comment: Comments.
- NE Pfam accession: Indicates a nested domain.
- NL Location: Location of nested domains - sequence ID, start and end of insert.
-
- Obsolete fields:
- -----------
- AL Alignment method of seed: The method used to align the seed members.
+ /*
+ * Generic per-File annotation, free text Magic features: #=GF NH
+ * #=GF TN Pfam descriptions: 7. DESCRIPTION OF FIELDS
+ *
+ * Compulsory fields: ------------------
+ *
+ * AC Accession number: Accession number in form PFxxxxx.version or
+ * PBxxxxxx. ID Identification: One word name for family. DE
+ * Definition: Short description of family. AU Author: Authors of the
+ * entry. SE Source of seed: The source suggesting the seed members
+ * belong to one family. GA Gathering method: Search threshold to
+ * build the full alignment. TC Trusted Cutoff: Lowest sequence score
+ * and domain score of match in the full alignment. NC Noise Cutoff:
+ * Highest sequence score and domain score of match not in full
+ * alignment. TP Type: Type of family -- presently Family, Domain,
+ * Motif or Repeat. SQ Sequence: Number of sequences in alignment. AM
+ * Alignment Method The order ls and fs hits are aligned to the model
+ * to build the full align. // End of alignment.
+ *
+ * Optional fields: ----------------
+ *
+ * DC Database Comment: Comment about database reference. DR Database
+ * Reference: Reference to external database. RC Reference Comment:
+ * Comment about literature reference. RN Reference Number: Reference
+ * Number. RM Reference Medline: Eight digit medline UI number. RT
+ * Reference Title: Reference Title. RA Reference Author: Reference
+ * Author RL Reference Location: Journal location. PI Previous
+ * identifier: Record of all previous ID lines. KW Keywords: Keywords.
+ * CC Comment: Comments. NE Pfam accession: Indicates a nested domain.
+ * NL Location: Location of nested domains - sequence ID, start and
+ * end of insert.
+ *
+ * Obsolete fields: ----------- AL Alignment method of seed: The
+ * method used to align the seed members.
*/
- // Let's save the annotations, maybe we'll be able to do something with them later...
+ // Let's save the annotations, maybe we'll be able to do something
+ // with them later...
Regex an = new Regex("(\\w+)\\s*(.*)");
if (an.search(annContent))
{
- alAnn.put(an.stringMatched(1), an.stringMatched(2));
+ if (an.stringMatched(1).equals("NH"))
+ {
+ treeString.append(an.stringMatched(2));
+ }
+ else if (an.stringMatched(1).equals("TN"))
+ {
+ if (treeString.length() > 0)
+ {
+ if (treeName == null)
+ {
+ treeName = "Tree " + (getTreeCount() + 1);
+ }
+ addNewickTree(treeName, treeString.toString());
+ }
+ treeName = an.stringMatched(2);
+ treeString = new StringBuffer();
+ }
+ setAlignmentProperty(an.stringMatched(1), an.stringMatched(2));
}
}
else if (annType.equals("GS"))
{
// Generic per-Sequence annotation, free text
- /* Pfam uses these features:
- Feature Description
- --------------------- -----------
- AC ACcession number
- DE DEscription
- DR ; ; Database Reference
- OS OrganiSm (species)
- OC Organism Classification (clade, etc.)
- LO Look (Color, etc.)
+ /*
+ * Pfam uses these features: Feature Description ---------------------
+ * ----------- AC ACcession number DE
+ * DEscription DR ; ; Database Reference OS
+ * OrganiSm (species) OC Organism Classification (clade, etc.)
+ * LO Look (Color, etc.)
*/
if (s.search(annContent))
{
String acc = s.stringMatched(1);
String type = s.stringMatched(2);
String content = s.stringMatched(3);
-
+ // TODO: store DR in a vector.
+ // TODO: store AC according to generic file db annotation.
Hashtable ann;
if (seqAnn.containsKey(acc))
{
@@ -299,37 +358,45 @@ public class StockholmFile
}
else if (annType.equals("GC"))
{
- System.out.println(annContent);
// Generic per-Column annotation, exactly 1 char per column
+ // always need a label.
+ if (x.search(annContent))
+ {
+ // parse out and create alignment annotation directly.
+ parseAnnotationRow(annotations, x.stringMatched(1),
+ x.stringMatched(2));
+ }
}
else if (annType.equals("GR"))
{
- // Generic per-Sequence AND per-Column markup, exactly 1 char per column
+ // Generic per-Sequence AND per-Column markup, exactly 1 char per
+ // column
/*
- Feature Description Markup letters
- ------- ----------- --------------
- SS Secondary Structure [HGIEBTSCX]
- SA Surface Accessibility [0-9X]
- (0=0%-10%; ...; 9=90%-100%)
- TM TransMembrane [Mio]
- PP Posterior Probability [0-9*]
- (0=0.00-0.05; 1=0.05-0.15; *=0.95-1.00)
- LI LIgand binding [*]
- AS Active Site [*]
- IN INtron (in or after) [0-2]
+ * Feature Description Markup letters ------- -----------
+ * -------------- SS Secondary Structure [HGIEBTSCX] SA Surface
+ * Accessibility [0-9X] (0=0%-10%; ...; 9=90%-100%) TM TransMembrane
+ * [Mio] PP Posterior Probability [0-9*] (0=0.00-0.05; 1=0.05-0.15;
+ * *=0.95-1.00) LI LIgand binding [*] AS Active Site [*] IN INtron (in
+ * or after) [0-2]
*/
if (s.search(annContent))
{
String acc = s.stringMatched(1);
String type = s.stringMatched(2);
- String seq = s.stringMatched(3);
- String description = new String();
-
+ String seq = new String(s.stringMatched(3));
+ String description = null;
// Check for additional information about the current annotation
- if (x.search(seq))
+ // We use a simple string tokenizer here for speed
+ StringTokenizer sep = new StringTokenizer(seq, " \t");
+ description = sep.nextToken();
+ if (sep.hasMoreTokens())
{
- description = x.stringMatched(1);
- seq = x.stringMatched(2);
+ seq = sep.nextToken();
+ }
+ else
+ {
+ seq = description;
+ description = new String();
}
// sequence id with from-to fields
@@ -337,12 +404,12 @@ public class StockholmFile
// Get an object with all the annotations for this sequence
if (seqAnn.containsKey(acc))
{
- //logger.debug("Found annotations for " + acc);
+ // logger.debug("Found annotations for " + acc);
ann = (Hashtable) seqAnn.get(acc);
}
else
{
- //logger.debug("Creating new annotations holder for " + acc);
+ // logger.debug("Creating new annotations holder for " + acc);
ann = new Hashtable();
seqAnn.put(acc, ann);
}
@@ -351,12 +418,12 @@ public class StockholmFile
// Get an object with all the content for an annotation
if (ann.containsKey("features"))
{
- //logger.debug("Found features for " + acc);
+ // logger.debug("Found features for " + acc);
features = (Hashtable) ann.get("features");
}
else
{
- //logger.debug("Creating new features holder for " + acc);
+ // logger.debug("Creating new features holder for " + acc);
features = new Hashtable();
ann.put("features", features);
}
@@ -364,12 +431,13 @@ public class StockholmFile
Hashtable content;
if (features.containsKey(this.id2type(type)))
{
- //logger.debug("Found content for " + this.id2type(type));
+ // logger.debug("Found content for " + this.id2type(type));
content = (Hashtable) features.get(this.id2type(type));
}
else
{
- //logger.debug("Creating new content holder for " + this.id2type(type));
+ // logger.debug("Creating new content holder for " +
+ // this.id2type(type));
content = new Hashtable();
features.put(this.id2type(type), content);
}
@@ -383,91 +451,133 @@ public class StockholmFile
}
else
{
- throw new IOException("Error parsing " + line);
+ System.err
+ .println("Warning - couldn't parse sequence annotation row line:\n"
+ + line);
+ // throw new IOException("Error parsing " + line);
}
}
else
{
- throw new IOException("Unknown annotation detected: " + annType + " " +
- annContent);
+ throw new IOException("Unknown annotation detected: " + annType
+ + " " + annContent);
}
}
}
+ if (treeString.length() > 0)
+ {
+ if (treeName == null)
+ {
+ treeName = "Tree " + (1 + getTreeCount());
+ }
+ addNewickTree(treeName, treeString.toString());
+ }
}
- public static String print(SequenceI[] s)
- {
- return "not yet implemented";
- }
-
- public String print()
- {
- return print(getSeqsAsArray());
- }
-
- private String id2type(String id)
+ private AlignmentAnnotation parseAnnotationRow(Vector annotation,
+ String label, String annots)
{
- // GR ids
- if (id.equals("SS"))
- {
- return "secondary structure";
- }
- else if (id.equals("SA"))
- {
- return "surface accessibility";
- }
- else if (id.equals("TM"))
- {
- return "transmembrane";
- }
- else if (id.equals("PP"))
- {
- return "posterior probability";
- }
- else if (id.equals("LI"))
- {
- return "ligand binding";
- }
- else if (id.equals("AS"))
+ String type = (label.indexOf("_cons") == label.length() - 5) ? label
+ .substring(0, label.length() - 5) : label;
+ boolean ss = false;
+ type = id2type(type);
+ if (type.equals("secondary structure"))
{
- return "active site";
+ ss = true;
}
- else if (id.equals("IN"))
+ // decide on secondary structure or not.
+ Annotation[] els = new Annotation[annots.length()];
+ for (int i = 0; i < annots.length(); i++)
{
- return "intron";
- }
- else if (id.equals("IR"))
- {
- return "interacting residue";
- }
- // GS ids
- else if (id.equals("AC"))
- {
- return "accession";
- }
- else if (id.equals("OS"))
- {
- return "organism";
+ String pos = annots.substring(i, i + 1);
+ Annotation ann;
+ ann = new Annotation(pos, "", ' ', 0f); // 0f is 'valid' null - will not
+ // be written out
+ if (ss)
+ {
+ ann.secondaryStructure = jalview.schemes.ResidueProperties
+ .getDssp3state(pos).charAt(0);
+ if (ann.secondaryStructure == pos.charAt(0) || pos.charAt(0) == 'C')
+ {
+ ann.displayCharacter = ""; // null; // " ";
+ }
+ else
+ {
+ ann.displayCharacter = " " + ann.displayCharacter;
+ }
+ }
+
+ els[i] = ann;
}
- else if (id.equals("CL"))
+ AlignmentAnnotation annot = null;
+ Enumeration e = annotation.elements();
+ while (e.hasMoreElements())
{
- return "class";
+ annot = (AlignmentAnnotation) e.nextElement();
+ if (annot.label.equals(type))
+ break;
+ annot = null;
}
- else if (id.equals("DE"))
+ if (annot == null)
{
- return "description";
+ annot = new AlignmentAnnotation(type, type, els);
+ annotation.addElement(annot);
}
- else if (id.equals("DR"))
+ else
{
- return "reference";
+ Annotation[] anns = new Annotation[annot.annotations.length
+ + els.length];
+ System.arraycopy(annot.annotations, 0, anns, 0,
+ annot.annotations.length);
+ System.arraycopy(els, 0, anns, annot.annotations.length, els.length);
+ annot.annotations = anns;
}
- else if (id.equals("LO"))
+ return annot;
+ }
+
+ public static String print(SequenceI[] s)
+ {
+ return "not yet implemented";
+ }
+
+ public String print()
+ {
+ return print(getSeqsAsArray());
+ }
+
+ private static Hashtable typeIds = null;
+ static
+ {
+ if (typeIds == null)
{
- return "look";
+ typeIds = new Hashtable();
+ typeIds.put("SS", "secondary structure");
+ typeIds.put("SA", "surface accessibility");
+ typeIds.put("TM", "transmembrane");
+ typeIds.put("PP", "posterior probability");
+ typeIds.put("LI", "ligand binding");
+ typeIds.put("AS", "active site");
+ typeIds.put("IN", "intron");
+ typeIds.put("IR", "interacting residue");
+ typeIds.put("AC", "accession");
+ typeIds.put("OS", "organism");
+ typeIds.put("CL", "class");
+ typeIds.put("DE", "description");
+ typeIds.put("DR", "reference");
+ typeIds.put("LO", "look");
+ typeIds.put("RF", "reference positions");
+
}
- else
+ }
+
+ private String id2type(String id)
+ {
+ if (typeIds.containsKey(id))
{
- return null;
+ return (String) typeIds.get(id);
}
+ System.err.println("Warning : Unknown Stockholm annotation type code "
+ + id);
+ return id;
}
}