diff options
Diffstat (limited to 'server.py')
-rwxr-xr-x | server.py | 82 |
1 files changed, 58 insertions, 24 deletions
@@ -370,8 +370,16 @@ def search(): def tableview(): with open("json_nodes.txt") as jsonfile: jnodes = json.load(jsonfile) - with open("json_edges.txt") as edgesjsonfile: - jedges = json.load(edgesjsonfile) + jedges ='' + file_edges = open('json_edges.txt', 'r') + for line in file_edges.readlines(): + if ':' not in line: + nodata_temp = 1 + else: + nodata_temp = 0 + with open("json_edges.txt") as edgesjsonfile: + jedges = json.load(edgesjsonfile) + break genename=session['query'] if len(genename)>3: genename = genename[0:3] @@ -384,14 +392,22 @@ def tableview(): gene_name = gene_name+added num_gene = gene_name.count(',')+1 message3="<b> Notes: </b><li> Click on the abstract count to read sentences linking the keyword and the gene. <li> Click on a gene to search its relations with top 200 addiction genes. <li> Click on a keyword to see the terms included in the search. <li>View the results in <a href='cytoscape'><b> a graph.</b></a>" - return render_template('tableview.html', num_gene=num_gene,session_path = session['path'], jedges=jedges, jnodes=jnodes,gene_name=gene_name, message3=message3) + return render_template('tableview.html', nodata_temp=nodata_temp, num_gene=num_gene,session_path = session['path'], jedges=jedges, jnodes=jnodes,gene_name=gene_name, message3=message3) @app.route("/tableview0") def tableview0(): with open("json_nodes.txt") as jsonfile: jnodes = json.load(jsonfile) - with open("json_edges.txt") as edgesjsonfile: - jedges = json.load(edgesjsonfile) + jedges ='' + file_edges = open('json_edges.txt', 'r') + for line in file_edges.readlines(): + if ':' not in line: + nodata_temp = 1 + else: + nodata_temp = 0 + with open("json_edges.txt") as edgesjsonfile: + jedges = json.load(edgesjsonfile) + break genename=session['query'] if len(genename)>3: genename = genename[0:3] @@ -403,8 +419,8 @@ def tableview0(): gene_name=gene_name.replace("'","") gene_name = gene_name+added num_gene = gene_name.count(',')+1 - message4="<b> Notes: </b><li> These are the words that have <b>zero</b> abstract counts. <li>View all the results in <a href='cytoscape'><b> a graph.</b></a>" - return render_template('tableview0.html', num_gene=num_gene,session_path = session['path'], jedges=jedges, jnodes=jnodes,gene_name=gene_name, message4=message4) + message4="<b> Notes: </b><li> These are the keywords that have <b>zero</b> abstract counts. <li>View all the results in <a href='cytoscape'><b> a graph.</b></a>" + return render_template('tableview0.html',nodata_temp=nodata_temp, num_gene=num_gene,session_path = session['path'], jedges=jedges, jnodes=jnodes,gene_name=gene_name, message4=message4) @app.route("/userarchive") def userarchive(): @@ -452,6 +468,7 @@ def date(): if ('email' in session): time_extension = str(select_date) time_extension = time_extension.split('_0_')[0] + gene_name1 = str(select_date).split('_0_')[1] time_extension = time_extension.replace(':', '_') time_extension = time_extension.replace('-', '_') session['path'] = tf_path+"/"+str(session['email'])+"/"+select_date+"/"+time_extension @@ -469,24 +486,41 @@ def date(): temp_file.write(line) with open("json_nodes.txt", "r") as jsonfile: jnodes = json.load(jsonfile) - with open("json_edges.txt", "r") as edgesjsonfile: - jedges = json.load(edgesjsonfile) + + jedges ='' + file_edges = open('json_edges.txt', 'r') + for line in file_edges.readlines(): + if ':' not in line: + nodata_temp = 1 + else: + nodata_temp = 0 + with open("json_edges.txt") as edgesjsonfile: + jedges = json.load(edgesjsonfile) + break gene_list=[] - for p in jedges['data']: - if p['source'] not in gene_list: - gene_list.append(p['source']) - if len(gene_list)>3: - gene_list = gene_list[0:3] - added = ",..." - else: - added = "" - gene_name = str(gene_list)[1:] - gene_name=gene_name[:-1] - gene_name=gene_name.replace("'","") - gene_name = gene_name+added - num_gene = gene_name.count(',')+1 + if nodata_temp == 0: + for p in jedges['data']: + if p['source'] not in gene_list: + gene_list.append(p['source']) + if len(gene_list)>3: + gene_list = gene_list[0:3] + added = ",..." + else: + added = "" + gene_name = str(gene_list)[1:] + gene_name=gene_name[:-1] + gene_name=gene_name.replace("'","") + gene_name = gene_name+added + num_gene = gene_name.count(',')+1 + else: + gene_name1 = gene_name1.replace("_", ", ") + gene_name = gene_name1 + num_gene = gene_name1.count(',')+1 + for i in range(0,num_gene): + gene_list.append(gene_name1.split(',')[i]) + session['query'] = gene_list message3="<b> Notes: </b><li>Click on the keywords to see the indicated number of abstracts <li> Click on a gene to search its relations with top 200 addiction genes<li>Click on a keyword to see the terms included in the search<li>Hover your pointer over a node will hide other links <li>Nodes can be moved around for better visibility, reload the page will restore the original layout<li> View the results in <a href='cytoscape'><b>a graph.</b></a>" - return render_template('tableview.html', title='', date=select_date, num_gene=num_gene,session_path = session['path'], jedges=jedges, jnodes=jnodes,gene_name=gene_name, message3=message3) + return render_template('tableview.html', title='',nodata_temp=nodata_temp, date=select_date, num_gene=num_gene,session_path = session['path'], jedges=jedges, jnodes=jnodes,gene_name=gene_name, message3=message3) @app.route('/cytoscape') def cytoscape(): @@ -515,7 +549,7 @@ def sentences(): out3+= "<li> "+ text + " <a href=\"https://www.ncbi.nlm.nih.gov/pubmed/?term=" + pmid +"\" target=_new>PMID:"+pmid+"<br></a>" pmid_temp = pmid pmid_list.append(pmid+cat0) - out1="<h3>"+gene0 + " and " + cat0 + "</h3><hr>\n" + out1="<h3>"+gene0 + " and " + cat0 + "</h3>\n" if len(pmid_list)>1: out2 = str(len(pmid_list)) + ' sentences in ' + str(len(pmid_list)) + ' studies' "<hr>\n" else: |