aboutsummaryrefslogtreecommitdiff
path: root/wqflask
diff options
context:
space:
mode:
authorLei Yan2014-02-21 13:58:13 -0600
committerLei Yan2014-02-21 13:58:13 -0600
commita48452cab16c1c9bd1c3efc476be4ae43d756707 (patch)
tree7af64e9bcce7744dce21882b0d80ae69815754c2 /wqflask
parent0e98cdd21f3c68a2918c0a70bd85e5debb4399d5 (diff)
downloadgenenetwork2-a48452cab16c1c9bd1c3efc476be4ae43d756707.tar.gz
On branch master
Diffstat (limited to 'wqflask')
-rw-r--r--wqflask/maintenance/dataset/load_genotypes.py207
1 files changed, 207 insertions, 0 deletions
diff --git a/wqflask/maintenance/dataset/load_genotypes.py b/wqflask/maintenance/dataset/load_genotypes.py
new file mode 100644
index 00000000..e4988446
--- /dev/null
+++ b/wqflask/maintenance/dataset/load_genotypes.py
@@ -0,0 +1,207 @@
+# GeneNetwork maintenance script
+# Load genotypes from geno file into database
+
+# Author: Lei Yan
+# Create Date: 2014-01-08
+# Last Update Date: 2014-01-10
+
+# import
+import sys
+import os
+import re
+import MySQLdb
+import ConfigParser
+
+def main(argv):
+
+ # load configuration from configuration file
+ config = ConfigParser.ConfigParser()
+ config.read(argv[1])
+ speciesid = config.get('configuration', 'speciesId')
+ inbredsetid = config.get('configuration', 'inbredsetid')
+ genofreezeid = config.get('configuration', 'genofreezeid')
+ genofile = config.get('configuration', 'genofile')
+ print "[configuration]\nspeciesid: %s\ninbredsetid: %s\ngenofreezeid: %s\ngenofile: %s\n" % (speciesid, inbredsetid, genofreezeid, genofile)
+
+ # variables
+ metadic = {}
+
+ # parse genofile
+ file_geno = open(genofile, 'r')
+ for line in file_geno:
+ line = line.strip()
+ if line.startswith('#'):
+ continue
+ if line.startswith('@'):
+ line = line.strip('@')
+ items = line.split(';')
+ for item in items:
+ kv = re.split(':|=', item)
+ metadic[kv[0].strip()] = kv[1].strip()
+ continue
+ if line.startswith("Chr"):
+ print "[meta dictionary]"
+ for k,v in metadic.items():
+ print "%s: %s" % (k, v)
+ print
+ continue
+ cells = line.split()
+ Chr = cells[0]
+ Locus = cells[1]
+ cM = cells[2]
+ Mb = cells[3]
+ print len(cells)
+ ?
+ return
+
+ # open db
+ host = 'localhost'
+ user = 'webqtl'
+ passwd = 'webqtl'
+ db = 'db_webqtl'
+ con = MySQLdb.Connect(db=db, user=user, passwd=passwd, host=host)
+ cursor = con.cursor()
+ # var
+ speciesid = int(argv[2])
+ inbredsetid = int(argv[3])
+ genofreezeid = int(argv[4])
+ sql = """
+ SELECT Id
+ FROM GenoData
+ ORDER BY Id DESC
+ LIMIT 1
+ """
+ cursor.execute(sql)
+ results = cursor.fetchall()
+ dataid = results[0][0]
+ print "speciesid: %s" % (speciesid)
+ print "inbredsetid: %s" % (inbredsetid)
+ print "genofreezeid: %s" % (genofreezeid)
+ print "dataid start: %s" % (dataid+1)
+ # samples
+ line = file_geno.readline()
+ sample_names = line.split()[4:]
+ sample_ids = []
+ print "get %d samples from file:\n%s" % (len(sample_names), sample_names)
+ for sample_name in sample_names:
+ sql = """
+ select Id
+ from Strain
+ where SpeciesId=%s
+ and Name like %s
+ """
+ cursor.execute(sql, (speciesid, sample_name))
+ results = cursor.fetchall()
+ if results:
+ sample_ids.append(results[0][0])
+ else:
+ print "insert sample %s" % (sample_name)
+ sql = """
+ INSERT INTO Strain
+ SET
+ SpeciesId=%s,
+ Name=%s,
+ Name2=%s
+ """
+ cursor.execute(sql, (speciesid, sample_name, sample_name))
+ sampleid = con.insert_id()
+ sample_ids.append(sampleid)
+ #
+ sql = """
+ SELECT OrderId
+ FROM StrainXRef
+ where InbredSetId=%s
+ ORDER BY OrderId DESC
+ LIMIT 1
+ """
+ cursor.execute(sql, (inbredsetid))
+ results = cursor.fetchall()
+ orderid = results[0][0] + 1
+ #
+ sql = """
+ INSERT INTO StrainXRef
+ SET
+ InbredSetId=%s,
+ StrainId=%s,
+ OrderId=%s,
+ Used_for_mapping=%s
+ """
+ cursor.execute(sql, (inbredsetid, sampleid, orderid, "N"))
+ print "load %d samples from DB:" % (len(sample_names))
+ for i in range(len(sample_names)):
+ print "%s\t%s" % (sample_names[i], sample_ids[i])
+ # parse geno file
+ index = 0
+ for line in file_geno:
+ index += 1
+ if index % 1000 == 0:
+ print index
+ items = line.split()
+ chr = items[0]
+ name = items[1]
+ cm = items[2]
+ mb = items[3]
+ values = items[4:]
+ # geno
+ sql = """
+ SELECT Id
+ FROM Geno
+ WHERE SpeciesId=%s
+ AND Name like %s
+ """
+ cursor.execute(sql, (speciesid, name))
+ results = cursor.fetchall()
+ if results:
+ genoid = results[0][0]
+ else:
+ print "insert geno %s" % (name)
+ sql = """
+ INSERT INTO Geno
+ SET
+ SpeciesId=%s,
+ Name=%s,
+ Marker_Name=%s,
+ Chr=%s,
+ Mb=%s
+ """
+ cursor.execute(sql, (speciesid, name, name, chr, mb))
+ genoid = con.insert_id()
+ # genodata
+ dataid += 1
+ for i in range(len(values)):
+ sample_id = sample_ids[i]
+ try:
+ value = int(values[i])
+ except ValueError:
+ continue
+ if not value in [-1, 0, 1]:
+ print sample_id, value
+ continue
+ sql = """
+ INSERT INTO GenoData
+ SET
+ Id=%s,
+ StrainId=%s,
+ value=%s
+ """
+ cursor.execute(sql, (dataid, sample_id, value))
+ # genoxref
+ sql = """
+ INSERT INTO GenoXRef
+ SET
+ GenoFreezeId=%s,
+ GenoId=%s,
+ DataId=%s,
+ cM=%s,
+ Used_for_mapping=%s
+ """
+ cursor.execute(sql, (genofreezeid, genoid, dataid, cm, 'N'))
+ print "Insert %d genoxref" % (index)
+ # close
+ file_geno.close()
+ con.close()
+
+# main
+if __name__ == "__main__":
+ main(sys.argv)
+ print "exit successfully"