From 45b6e6a5eea13dbd06045e8307e371994b8c3083 Mon Sep 17 00:00:00 2001 From: Zachary Sloan Date: Tue, 30 Oct 2012 18:22:27 -0500 Subject: Worked on search_result.py code related to adding clause item to queries Continued work on redesigning the trait data page design --- wqflask/base/webqtlTrait.py | 12 +- wqflask/wqflask/search_results.py | 249 ++++++++++++++++------------- wqflask/wqflask/templates/show_trait.html | 253 ++++-------------------------- 3 files changed, 177 insertions(+), 337 deletions(-) (limited to 'wqflask') diff --git a/wqflask/base/webqtlTrait.py b/wqflask/base/webqtlTrait.py index 46af6683..51d36ab2 100755 --- a/wqflask/base/webqtlTrait.py +++ b/wqflask/base/webqtlTrait.py @@ -226,7 +226,7 @@ class webqtlTrait: def retrieveData(self, samplelist=None): - + if samplelist == None: samplelist = [] assert self.db and self.cursor @@ -331,7 +331,7 @@ class webqtlTrait: self.cursor.execute(query) results = self.cursor.fetchall() self.data.clear() - + if results: self.mysqlid = results[0][-1] #if samplelist: @@ -345,7 +345,7 @@ class webqtlTrait: name = item[0] self.data[name] = webqtlCaseData(*item) #name, value, variance, num_cases) #end for - # else: + # else: # for item in results: # val = item[1] # if val != None: @@ -604,7 +604,7 @@ class webqtlTrait: formatted += "; " + self.probe_target_description else: formatted = "Not available" - return formatted + return formatted.capitalize() @property def alias_fmt(self): @@ -660,7 +660,7 @@ class webqtlTrait: else: return dict(name = self.db.fullname, url = webqtlConfig.INFOPAGEHREF % self.db.name) - + def calculate_correlation(self, values, method): """Calculate the correlation value and p value according to the method specified""" @@ -693,4 +693,4 @@ class webqtlTrait: else: ZValue = 0.5*log((1.0+self.correlation)/(1.0-self.correlation)) ZValue = ZValue*sqrt(self.overlap-3) - self.p_value = 2.0*(1.0 - reaper.normp(abs(ZValue))) + self.p_value = 2.0*(1.0 - reaper.normp(abs(ZValue))) diff --git a/wqflask/wqflask/search_results.py b/wqflask/wqflask/search_results.py index a2b0596c..29a497fa 100644 --- a/wqflask/wqflask/search_results.py +++ b/wqflask/wqflask/search_results.py @@ -365,114 +365,111 @@ class SearchResultPage(templatePage): #last_result = False self.trait_list = [] - for result in self.results: - if not result: - continue - #last_result = False - - #for item in result: - print("foo locals are:", locals()) - probe_set_id = result[1] - print("probe_set_id is:", pf(probe_set_id)) - this_trait = webqtlTrait(db=self.dataset, name=probe_set_id, cursor=self.cursor) - this_trait.retrieveInfo(QTL=True) - print("this_trait is:", pf(this_trait)) - self.trait_list.append(this_trait) - print("self.trait_list is:", pf(self.trait_list)) - - ############## - # Excel file # - ############## - - # Todo: Replace this with official Python temp file naming functions? - filename= webqtlUtil.genRandStr("Search_") - #xlsUrl = HT.Input(type='button', value = 'Download Table', onClick= "location.href='/tmp/%s.xls'" % filename, Class='button') - # Create a new Excel workbook - #workbook = xl.Writer('%s.xls' % (webqtlConfig.TMPDIR+filename)) - #headingStyle = workbook.add_format(align = 'center', bold = 1, border = 1, size=13, fg_color = 0x1E, color="white") - - #XZ, 3/18/2010: pay attention to the line number of header in this file. As of today, there are 7 lines. - #worksheet = self.createExcelFileWithTitleAndFooter(workbook=workbook, db=this_trait.db, returnNumber=len(self.trait_list)) - newrow = 7 - - #### Excel file stuff stops + # result_set represents the results for each search term; a search of + # "shh grin2b" would have two sets of results, one for each term + for result_set in self.results: + for result in result_set: + if not result: + continue + #last_result = False - #tbl = HT.TableLite(cellSpacing=2,cellPadding=0,width="90%",border=0) - #seq = self.pageNumber*self.NPerPage+1 //Edited out because we show all results in one page now - Zach 2/22/11 - seq = 1 - group = self.dataset.group - self.form_name = form_name = 'show_dataset_'+group + seq = 1 + group = self.dataset.group + self.form_name = form_name = 'show_dataset_'+group - tblobj = {} - species = webqtlDatabaseFunction.retrieveSpecies(cursor=self.cursor, RISet=group) + tblobj = {} + species = webqtlDatabaseFunction.retrieveSpecies(cursor=self.cursor, RISet=group) - if this_trait.db.type=="Geno": - tblobj['header'] = self.getTableHeaderForGeno(worksheet=worksheet, newrow=newrow, headingStyle=headingStyle) + #### Excel file - newrow += 1 - sortby = self.getSortByValue(datasetType="Geno") + # Todo: Replace this with official Python temp file naming functions? + filename= webqtlUtil.genRandStr("Search_") + #xlsUrl = HT.Input(type='button', value = 'Download Table', onClick= "location.href='/tmp/%s.xls'" % filename, Class='button') + # Create a new Excel workbook + #workbook = xl.Writer('%s.xls' % (webqtlConfig.TMPDIR+filename)) + #headingStyle = workbook.add_format(align = 'center', bold = 1, border = 1, size=13, fg_color = 0x1E, color="white") - #tblobj['body'] = self.getTableBodyForGeno(trait_list=self.trait_list, form_name=form_name, worksheet=worksheet, newrow=newrow) + #XZ, 3/18/2010: pay attention to the line number of header in this file. As of today, there are 7 lines. + #worksheet = self.createExcelFileWithTitleAndFooter(workbook=workbook, db=this_trait.db, returnNumber=len(self.trait_list)) + newrow = 7 - #workbook.close() - #objfile = open('%s.obj' % (webqtlConfig.TMPDIR+filename), 'wb') - #cPickle.dump(tblobj, objfile) - #objfile.close() + #### Excel file stuff stops - #div = HT.Div(webqtlUtil.genTableObj(tblobj, filename, sortby), Id="sortable") - # - #pageTable.append(HT.TR(HT.TD(div))) + if self.dataset.type == "ProbeSet": + #for item in result: + print("foo locals are:", locals()) + probe_set_id = result[1] + print("probe_set_id is:", pf(probe_set_id)) + this_trait = webqtlTrait(db=self.dataset, name=probe_set_id, cursor=self.cursor) + this_trait.retrieveInfo(QTL=True) + print("this_trait is:", pf(this_trait)) + self.trait_list.append(this_trait) + print("self.trait_list is:", pf(self.trait_list)) - elif this_trait.db.type=="Publish": - #tblobj['header'] = self.getTableHeaderForPublish(worksheet=worksheet, newrow=newrow, headingStyle=headingStyle) + #tblobj['header'] = self.getTableHeaderForProbeSet(worksheet=worksheet, newrow=newrow, headingStyle=headingStyle) - #newrow += 1 + #newrow += 1 - sortby = self.getSortByValue(datasetType="Publish") + sortby = self.getSortByValue(datasetType="ProbeSet") - #tblobj['body'] = self.getTableBodyForPublish(trait_list=self.trait_list, formName=mainfmName, worksheet=worksheet, newrow=newrow, species=species) + tblobj['body'] = self.getTableBodyForProbeSet(trait_list=self.trait_list, formName=self.form_name, newrow=newrow, species=species) - #workbook.close() - #objfile = open('%s.obj' % (webqtlConfig.TMPDIR+filename), 'wb') - #cPickle.dump(tblobj, objfile) - #objfile.close() + #workbook.close() + #objfile = open('%s.obj' % (webqtlConfig.TMPDIR+filename), 'wb') + #cPickle.dump(tblobj, objfile) + #objfile.close() - #div = HT.Div(webqtlUtil.genTableObj(tblobj, filename, sortby), Id="sortable") + #div = HT.Div(webqtlUtil.genTableObj(tblobj, filename, sortby), Id="sortable") - #pageTable.append(HT.TR(HT.TD(div))) + #pageTable.append(HT.TR(HT.TD(div))) + elif self.dataset.type == "Publish": + #tblobj['header'] = self.getTableHeaderForPublish(worksheet=worksheet, newrow=newrow, headingStyle=headingStyle) - elif this_trait.db.type=="ProbeSet": - #tblobj['header'] = self.getTableHeaderForProbeSet(worksheet=worksheet, newrow=newrow, headingStyle=headingStyle) + #newrow += 1 - #newrow += 1 + sortby = self.getSortByValue(datasetType="Publish") - sortby = self.getSortByValue(datasetType="ProbeSet") + #tblobj['body'] = self.getTableBodyForPublish(trait_list=self.trait_list, formName=mainfmName, worksheet=worksheet, newrow=newrow, species=species) - tblobj['body'] = self.getTableBodyForProbeSet(trait_list=self.trait_list, formName=self.form_name, newrow=newrow, species=species) - # - #workbook.close() - #objfile = open('%s.obj' % (webqtlConfig.TMPDIR+filename), 'wb') - #cPickle.dump(tblobj, objfile) - #objfile.close() + #workbook.close() + #objfile = open('%s.obj' % (webqtlConfig.TMPDIR+filename), 'wb') + #cPickle.dump(tblobj, objfile) + #objfile.close() - #div = HT.Div(webqtlUtil.genTableObj(tblobj, filename, sortby), Id="sortable") + #div = HT.Div(webqtlUtil.genTableObj(tblobj, filename, sortby), Id="sortable") - #pageTable.append(HT.TR(HT.TD(div))) + #pageTable.append(HT.TR(HT.TD(div))) + elif self.dataset.type == "Geno": + tblobj['header'] = self.getTableHeaderForGeno(worksheet=worksheet, newrow=newrow, headingStyle=headingStyle) + newrow += 1 + sortby = self.getSortByValue(datasetType="Geno") - #traitForm = HT.Form(cgi= os.path.join(webqtlConfig.CGIDIR, webqtlConfig.SCRIPTFILE), enctype='multipart/form-data', name=thisFormName, submit=HT.Input(type='hidden')) - hddn = {'FormID':'showDatabase','ProbeSetID':'_','database':'_','CellID':'_','group':group} - hddn['incparentsf1']='ON' - # for key in hddn.keys(): - # traitForm.append(HT.Input(name=key, value=hddn[key], type='hidden')) - # - # traitForm.append(HT.P(),pageTable) - # - # TD_LR.append(traitForm) - # if len(self.results) > 1 and i < len(self.results) - 1: - # last_result = True - #if last_result: - # TD_LR.contents.pop() + #tblobj['body'] = self.getTableBodyForGeno(trait_list=self.trait_list, form_name=form_name, worksheet=worksheet, newrow=newrow) + + #workbook.close() + #objfile = open('%s.obj' % (webqtlConfig.TMPDIR+filename), 'wb') + #cPickle.dump(tblobj, objfile) + #objfile.close() + + #div = HT.Div(webqtlUtil.genTableObj(tblobj, filename, sortby), Id="sortable") + # + #pageTable.append(HT.TR(HT.TD(div))) + + + #traitForm = HT.Form(cgi= os.path.join(webqtlConfig.CGIDIR, webqtlConfig.SCRIPTFILE), enctype='multipart/form-data', name=thisFormName, submit=HT.Input(type='hidden')) + hddn = {'FormID':'showDatabase','ProbeSetID':'_','database':'_','CellID':'_','group':group} + hddn['incparentsf1']='ON' + # for key in hddn.keys(): + # traitForm.append(HT.Input(name=key, value=hddn[key], type='hidden')) + # + # traitForm.append(HT.P(),pageTable) + # + # TD_LR.append(traitForm) + # if len(self.results) > 1 and i < len(self.results) - 1: + # last_result = True + #if last_result: + # TD_LR.contents.pop() def executeQuery(self): @@ -642,6 +639,8 @@ class SearchResultPage(templatePage): print("fd.search_terms:", self.fd['search_terms']) self.search_terms = parser.parse(self.fd['search_terms']) print("After parsing:", self.search_terms) + + #print("ORkeyword is:", pf(self.ORkeyword)) #self.ANDkeyword2 = parser.parse(self.ANDkeyword) #self.ORkeyword2 = parser.parse(self.ORkeyword) @@ -662,11 +661,8 @@ class SearchResultPage(templatePage): ###remove remain parethesis, could be input with syntax error #self.ORkeyword2 = re.sub(re.compile('\s*\([\s\S]*\)'), '', self.ORkeyword2) #self.ORkeyword2 = self.encregexp(self.ORkeyword2) - - #if self.search_terms: - #full_text = [] - #ANDFulltext = [] - #ORFulltext = [] + + self.results = [] for item in self.search_terms: search_term = item['search_term'] # self.nkeywords += 1 @@ -683,8 +679,19 @@ class SearchResultPage(templatePage): # fulltext = ORFulltext print("item is:", pf(search_term)) - + + + clause_item = ( +""" MATCH (ProbeSet.Name, + ProbeSet.description, + ProbeSet.symbol, + alias, + GenbankId, + UniGeneId, + Probe_Target_Description) + AGAINST ('%s' IN BOOLEAN MODE) """ % self.db_conn.escape_string(search_term)) if self.dataset.type == "ProbeSet": + query = ( """SELECT distinct 0, ProbeSet.Name as TNAME, 0 as thistable, ProbeSetXRef.Mean as TMEAN, @@ -693,31 +700,55 @@ class SearchResultPage(templatePage): ProbeSet.Chr_num as TCHR_NUM, ProbeSet.Mb as TMB, ProbeSet.Symbol as TSYMBOL, - ProbeSet.name_num as TNAME_NUM FROM ProbeSetXRef, - ProbeSet WHERE (MATCH (ProbeSet.Name, ProbeSet.description, ProbeSet.symbol, - alias, GenbankId, UniGeneId, Probe_Target_Description) - AGAINST ('%s' IN BOOLEAN MODE)) - and ProbeSet.Id = ProbeSetXRef.ProbeSetId - and ProbeSetXRef.ProbeSetFreezeId = %s + ProbeSet.name_num as TNAME_NUM + FROM ProbeSetXRef, ProbeSet + WHERE %s + and ProbeSet.Id = ProbeSetXRef.ProbeSetId + and ProbeSetXRef.ProbeSetFreezeId = %s """ % (self.db_conn.escape_string(search_term), + self.db_conn.escape_string(clause_item), self.db_conn.escape_string(str(self.dataset.id)))) + + elif self.dataset.type == "Publish": + include_geno = "" + if search_term.find("Geno.name") >= 0: + include_geno = " Geno, " + + query = ( +"""SELECT 0, PublishXRef.Id, PublishFreeze.createtime as thistable, + Publication.PubMed_ID as Publication_PubMed_ID, + Phenotype.Post_publication_description as Phenotype_Name + FROM %s PublishFreeze, Publication, PublishXRef, Phenotype + WHERE PublishXRef.InbredSetId = %s and %s and + PublishXRef.PhenotypeId = Phenotype.Id and + PublishXRef.PublicationId = Publication.Id and + PublishFreeze.Id = %s + """ % (include_geno, + self.db_conn.escape_string(str(self.dataset.group_id)), + self.db_conn.escape_string(clause_item), + self.db_conn.escape_string(str(self.dataset.id)))) + + elif self.dataset.type == "Geno": + query = ( +"""SELECT 0, Geno.Name, GenoFreeze.createtime as thistable, + Geno.Name as Geno_Name, + Geno.Source2 as Geno_Source2, + Geno.chr_num as Geno_chr_num, + Geno.Mb as Geno_Mb + FROM GenoXRef, GenoFreeze, Geno + WHERE %s and Geno.Id = GenoXRef.GenoId and + GenoXRef.GenoFreezeId = GenoFreeze.Id and + GenoFreeze.Id = %d + """% (self.db_conn.escape_string(clause_item), + self.db_conn.escape_string(str(self.dataset.id)))) - self.cursor.execute(query) - #print("query is:", pf(self.query)) - #self.cursor.execute(self.query) - self.results = self.cursor.fetchall() + self.cursor.execute(query) + self.results.append(self.cursor.fetchall()) print("self.results is:", pf(self.results)) -#["(SELECT distinct 0, ProbeSet.Name as TNAME, 0 as thistable,\n\t\t\t\t\t\tProbeSetXRef.Mean as TMEAN, -# ProbeSetXRef.LRS as TLRS, ProbeSetXRef.PVALUE as TPVALUE,\n\t\t\t\t\t\tProbeSet.Chr_num as TCHR_NUM, -# Pr beSet.Mb as TMB, ProbeSet.Symbol as TSYMBOL,\n\t\t\t\t\t\tProbeSet.name_num as TNAME_NUM FROM -# ProbeSetXRef, ProbeSet \n\t\t\t\t\t\tWHERE ( MATCH (ProbeSet.Name,ProbeSet.description,ProbeSet.symbol, -# alias GenbankId, UniGeneId, Probe_Target_Description) -# AGAINST ('shh' IN BOOLEAN MODE) ) -# and ProbeSet.Id = ProbeSetXRef.ProbeSetId and ProbeSetXRef.ProbeSetFreezeId = 112\n\t\t\t\t\t\t)"] #if self.dataset.type == "ProbeSet" and search_term.find('.') < 0 and search_term.find('\'') < 0: diff --git a/wqflask/wqflask/templates/show_trait.html b/wqflask/wqflask/templates/show_trait.html index ab662a77..d7b81562 100644 --- a/wqflask/wqflask/templates/show_trait.html +++ b/wqflask/wqflask/templates/show_trait.html @@ -48,240 +48,49 @@
{{ "%i" % (this_trait.probe_set_blat_score) }}
- - - - -
-
- {# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - #} - {% for key in hddn %} - - {% endfor %} -
-
- Trait Data and Analysis  for Record ID 1441186_at -
-
- -

  Details and Links

- -

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - + + {% for key in hddn %} + + {% endfor %} - - - +
+
+ -
- - - - - - + - - - + + - - - - - - - -
Gene Symbol:{{ this_trait.symbol }}
Aliases:{{ this_trait.alias_fmt }}
Description:{{ this_trait.description_fmt }}
Location:{{ this_trait.location_fmt }}
Target Score: - - - BLAT specificity - : {{ "%.1f" % (this_trait.probe_set_specificity) }}    - Score: {{ "%i" % (this_trait.probe_set_blat_score) }}   - -
Species and Group:{{ this_trait.species.capitalize() }}, {{fd.RISet}}
Database: - {{ this_trait.database.name }} -
Resource Links: - - - - Gene - - -   UniGene  GenBank  HomoloGene  
UCSC  BioGPS  STRING  PANTHER  Gemma  SynDB  ABA  

- - - - - - - - - - - - - - - - - - - +
-
- + - + - + - + - + - + - - - -
-- -  Check probe locations at UCSC  Write or review comments about this gene -  View SNPs and Indels -  View probes, SNPs, and RNA-seq at UTHSC -  Check sequence of probes
AddFindVerifyGeneWikiSNPsRNA-seqProbes

  Basic Statistics

-- cgit v1.2.3