view tools/data_source/microbial_import_code.py @ 1:cdcb0ce84a1b

Uploaded
author xuebing
date Fri, 09 Mar 2012 19:45:15 -0500
parents 9071e359b9a3
children
line wrap: on
line source


def load_microbial_data( GALAXY_DATA_INDEX_DIR, sep='\t' ):
    # FIXME: this function is duplicated in the DynamicOptions class.  It is used here only to
    # set data.name in exec_after_process(). 
    microbe_info= {}
    orgs = {}
    
    filename = "%s/microbial_data.loc" % GALAXY_DATA_INDEX_DIR
    for i, line in enumerate( open( filename ) ):
        line = line.rstrip( '\r\n' )
        if line and not line.startswith( '#' ):
            fields = line.split( sep )
            #read each line, if not enough fields, go to next line
            try:
                info_type = fields.pop(0)
                if info_type.upper() == "ORG":
                    #ORG     12521   Clostridium perfringens SM101   bacteria        Firmicutes      CP000312,CP000313,CP000314,CP000315     http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=genomeprj&cmd=Retrieve&dopt=Overview&list_uids=12521
                    org_num = fields.pop(0)
                    name = fields.pop(0)
                    kingdom = fields.pop(0)
                    group = fields.pop(0)
                    chromosomes = fields.pop(0)
                    info_url = fields.pop(0)
                    link_site = fields.pop(0)
                    if org_num not in orgs:
                        orgs[ org_num ] = {}
                        orgs[ org_num ][ 'chrs' ] = {}
                    orgs[ org_num ][ 'name' ] = name
                    orgs[ org_num ][ 'kingdom' ] = kingdom
                    orgs[ org_num ][ 'group' ] = group
                    orgs[ org_num ][ 'chromosomes' ] = chromosomes
                    orgs[ org_num ][ 'info_url' ] = info_url
                    orgs[ org_num ][ 'link_site' ] = link_site
                elif info_type.upper() == "CHR":
                    #CHR     12521   CP000315        Clostridium perfringens phage phiSM101, complete genome 38092   110684521       CP000315.1
                    org_num = fields.pop(0)
                    chr_acc = fields.pop(0)
                    name = fields.pop(0)
                    length = fields.pop(0)
                    gi = fields.pop(0)
                    gb = fields.pop(0)
                    info_url = fields.pop(0)
                    chr = {}
                    chr[ 'name' ] = name
                    chr[ 'length' ] = length
                    chr[ 'gi' ] = gi
                    chr[ 'gb' ] = gb
                    chr[ 'info_url' ] = info_url
                    if org_num not in orgs:
                        orgs[ org_num ] = {}
                        orgs[ org_num ][ 'chrs' ] = {}
                    orgs[ org_num ][ 'chrs' ][ chr_acc ] = chr
                elif info_type.upper() == "DATA":
                    #DATA    12521_12521_CDS 12521   CP000315        CDS     bed     /home/djb396/alignments/playground/bacteria/12521/CP000315.CDS.bed
                    uid = fields.pop(0)
                    org_num = fields.pop(0)
                    chr_acc = fields.pop(0)
                    feature = fields.pop(0)
                    filetype = fields.pop(0)
                    path = fields.pop(0)
                    data = {}
                    data[ 'filetype' ] = filetype
                    data[ 'path' ] = path
                    data[ 'feature' ] = feature

                    if org_num not in orgs:
                        orgs[ org_num ] = {}
                        orgs[ org_num ][ 'chrs' ] = {}
                    if 'data' not in orgs[ org_num ][ 'chrs' ][ chr_acc ]:
                        orgs[ org_num ][ 'chrs' ][ chr_acc ][ 'data' ] = {}
                    orgs[ org_num ][ 'chrs' ][ chr_acc ][ 'data' ][ uid ] = data
                else: continue
            except: continue
    for org_num in orgs:
        org = orgs[ org_num ]
        if org[ 'kingdom' ] not in microbe_info:
            microbe_info[ org[ 'kingdom' ] ] = {}
        if org_num not in microbe_info[ org[ 'kingdom' ] ]:
            microbe_info[ org[ 'kingdom' ] ][org_num] = org
    return microbe_info

#post processing, set build for data and add additional data to history
from galaxy import datatypes, config, jobs, tools
from shutil import copyfile

def exec_after_process(app, inp_data, out_data, param_dict, tool, stdout, stderr):
    base_dataset = out_data.items()[0][1]
    history = base_dataset.history
    if history == None:
        print "unknown history!"
        return
    kingdom = param_dict.get( 'kingdom', None )
    #group = param_dict.get( 'group', None )
    org = param_dict.get( 'org', None )
    
    #if not (kingdom or group or org):
    if not (kingdom or org):
        print "Parameters are not available."
    #workflow passes galaxy.tools.parameters.basic.UnvalidatedValue instead of values
    if isinstance( kingdom, tools.parameters.basic.UnvalidatedValue ):
        kingdom = kingdom.value
    if isinstance( org, tools.parameters.basic.UnvalidatedValue ):
        org = org.value
    
    GALAXY_DATA_INDEX_DIR = app.config.tool_data_path
    microbe_info = load_microbial_data( GALAXY_DATA_INDEX_DIR, sep='\t' )
    new_stdout = ""
    split_stdout = stdout.split("\n")
    basic_name = ""
    for line in split_stdout:
        fields = line.split("\t")
        if fields[0] == "#File1":
            description = fields[1]
            chr = fields[2]
            dbkey = fields[3]
            file_type = fields[4]
            name, data = out_data.items()[0]
            data.set_size()
            basic_name = data.name
            data.name = data.name + " (" + microbe_info[kingdom][org]['chrs'][chr]['data'][description]['feature'] +" for " + microbe_info[kingdom][org]['name'] + ":" + chr + ")"
            data.dbkey = dbkey
            data.info = data.name
            data = app.datatypes_registry.change_datatype( data, file_type )
            data.init_meta()
            data.set_peek()
            app.model.context.add( data )
            app.model.context.flush()
        elif fields[0] == "#NewFile":
            description = fields[1]
            chr = fields[2]
            dbkey = fields[3]
            filepath = fields[4]
            file_type = fields[5]
            newdata = app.model.HistoryDatasetAssociation( create_dataset = True, sa_session = app.model.context ) #This import should become a library
            newdata.set_size()
            newdata.extension = file_type
            newdata.name = basic_name + " (" + microbe_info[kingdom][org]['chrs'][chr]['data'][description]['feature'] +" for "+microbe_info[kingdom][org]['name']+":"+chr + ")"
            app.model.context.add( newdata )
            app.model.context.flush()
            app.security_agent.copy_dataset_permissions( base_dataset.dataset, newdata.dataset )
            history.add_dataset( newdata )
            app.model.context.add( history )
            app.model.context.flush()
            try:
                copyfile(filepath,newdata.file_name)
                newdata.info = newdata.name
                newdata.state = jobs.JOB_OK
            except:
                newdata.info = "The requested file is missing from the system."
                newdata.state = jobs.JOB_ERROR
            newdata.dbkey = dbkey
            newdata.init_meta()
            newdata.set_peek()
            app.model.context.flush()