Source Code for Module TEES.ExampleBuilders.FeatureBuilders.DrugFeatureBuilder

  1  import sys, os 
  2  sys.path.append(os.path.join(os.path.dirname(os.path.abspath(__file__)), "../..")) 
  3  from Core.IdSet import IdSet 
  4  import Core.ExampleUtils as ExampleUtils 
  5  from FeatureBuilder import FeatureBuilder 
  6  import Utils.Settings as Settings 
  7  import Utils.Download 
  8  import Utils.ElementTreeUtils as ETUtils 
  9  from collections import defaultdict 
 10   
 11 -def installDrugBank(destPath=None, downloadPath=None, redownload=False, updateLocalSettings=False): 
 12      print >> sys.stderr, "---------------", "Downloading Drug Bank XML", "---------------" 
 13      print >> sys.stderr, "See http://www.drugbank.ca/downloads for conditions of use" 
 14      if destPath == None: 
 15          destPath = os.path.join(Settings.DATAPATH, "resources") 
 16      if downloadPath == None: 
 17          downloadPath = os.path.join(Settings.DATAPATH, "resources/download") 
 18      filenames = Utils.Download.downloadAndExtract(Settings.URL["DRUG_BANK_XML"], destPath, downloadPath, redownload=redownload) 
 19      assert len(filenames) == 1 
 20      Settings.setLocal("DRUG_BANK_XML", os.path.join(destPath, filenames[0]), updateLocalSettings) 
 21   
 22 -class DrugFeatureBuilder(FeatureBuilder): 
 23      data = None 
 24       
 25 -    def __init__(self, featureSet=None): 
 26          FeatureBuilder.__init__(self, featureSet) 
 27          if not hasattr(Settings, "DRUG_BANK_XML"): 
 28              print >> sys.stderr, "Drug Bank XML not installed, installing now" 
 29              installDrugBank(updateLocalSettings=True) 
 30          drugBankFile = Settings.DRUG_BANK_XML 
 31          #drugBankFile = "/home/jari/data/DDIExtraction2011/resources/drugbank.xml" 
 32          # Load drug data into memory on first call to constructor 
 33          if DrugFeatureBuilder.data == None: 
 34              DrugFeatureBuilder.data, DrugFeatureBuilder.nameToId = prepareDrugBank(drugBankFile) 
 35              DrugFeatureBuilder.interactionPairs = buildInteractionPairs(DrugFeatureBuilder.data) 
 36       
 37 -    def buildPairFeatures(self, e1, e2): 
 38          e1Name = normalizeDrugName(e1.get("text")) 
 39          e2Name = normalizeDrugName(e2.get("text")) 
 40          interactionType = self.getInteraction(e1Name, e2Name) 
 41          if self.getInteraction(e1Name, e2Name) == True: 
 42              self.setFeature("DrugBankPairTrueInt") 
 43          else: 
 44              self.setFeature("NotDrugInteraction") 
 45              if interactionType == "UNKNOWN_NAME": 
 46                  self.setFeature("PairNotInDrugBank") 
 47              else: 
 48                  self.setFeature("DrugBankPairFalseInt") 
 49       
 50 -    def getMTMXAttrs(self, e1, e2, attr): 
 51          rv = [str(e1.get(attr)).lower().replace(" ", ""), str(e2.get(attr)).lower().replace(" ", "")] 
 52          if rv[0] == "": rv[0] = "none" 
 53          if rv[1] == "": rv[1] = "none" 
 54          rv.sort() 
 55          return rv 
 56           
 57 -    def buildMTMXFeatures(self, e1, e2): 
 58          names = self.getMTMXAttrs(e1, e2, "mtmxName") 
 59          self.setFeature("mtmxNames-" + "-".join(names)) 
 60          if names[0] == names[1]: 
 61              if names[0] in ["", "none"]: 
 62                  self.setFeature("mtmxNames-both_unknown") 
 63              else: 
 64                  self.setFeature("mtmxNames-both_identical") 
 65          self.setFeature("mtmxShortNames-" + "-".join(self.getMTMXAttrs(e1, e2, "mtmxNameShort"))) 
 66          mtmxCuis = self.getMTMXAttrs(e1, e2, "mtmxCui") 
 67          for mtmxCui in mtmxCuis: 
 68              self.setFeature("mtmxCui_" + mtmxCui) 
 69          self.setFeature("mtmxCuis-" + "-".join(mtmxCuis)) 
 70          # Probabilities 
 71          rv = self.getMTMXAttrs(e1, e2, "mtmxProb") 
 72          if rv[0] in ["", "none"]: rv[0] = "0" 
 73          if rv[1] in ["", "none"]: rv[1] = "0" 
 74          rv[0] = int(rv[0]) 
 75          rv[1] = int(rv[1]) 
 76          assert rv[0] <= 1000 and rv[1] <= 1000, (rv[0], rv[1]) 
 77          rv.sort() 
 78          self.setFeature("mtmxProbMin", float(rv[0]) / 1000.0) 
 79          self.setFeature("mtmxProbMax", float(rv[1]) / 1000.0) 
 80          # Semtypes 
 81          sem = self.getMTMXAttrs(e1, e2, "mtmxSemTypes") 
 82          #print sem 
 83          for i in sem[0].split(","): 
 84              for j in sem[1].split(","): 
 85                  semPair = [i, j] 
 86                  semPair.sort() 
 87                  #print "semPair", semPair 
 88                  self.setFeature("semPair-" + "-".join(semPair)) 
 89                  self.setFeature("semType-" + i) 
 90                  self.setFeature("semType-" + j) 
 91       
 92 -    def getInteraction(self, e1Name, e2Name): 
 93          e1Name = normalizeDrugName(e1Name) 
 94          e2Name = normalizeDrugName(e2Name) 
 95          e1Ids = DrugFeatureBuilder.nameToId[e1Name] 
 96          e2Ids = DrugFeatureBuilder.nameToId[e2Name] 
 97          #print e1Ids, e2Ids 
 98          if len(e1Ids) == 0 or len(e2Ids) == 0: 
 99              return "UNKNOWN_NAME" # unknown drug name 
100          for id1 in e1Ids: 
101              for id2 in e2Ids: 
102                  if DrugFeatureBuilder.interactionPairs[id1][id2]: 
103                      return True 
104          return False 
105   
106 -def normalizeDrugName(text): 
107      return text.replace("-","").replace("/","").replace(",","").replace("\\","").replace(" ","").lower() 
108   
109 -def getNestedItems(parent, term, data, preTag, termPlural=None, verbose=False): 
110      if termPlural != None: 
111          items = parent.find(preTag+termPlural).findall(preTag+term) 
112      else: 
113          items = parent.find(preTag+term+"s").findall(preTag+term) 
114      for item in items: 
115          data[term].append(item.text) 
116          if verbose: print "  " + term + ": " + item.text 
117           
118 -def resolveInteractions(data, verbose=False): 
119      counts = defaultdict(int) 
120      if verbose: print "Resolving Interactions" 
121      for id in data: 
122          if verbose: print id, data[id]["name"] 
123          for interaction in data[id]["interaction"]: 
124              partnerDBId = str(interaction[0]) 
125              partnerDBId = "DB" + (5 - len(partnerDBId)) * "0" + partnerDBId 
126              interaction[0] = partnerDBId 
127              if partnerDBId in data: 
128                  interaction[1] = data[partnerDBId]["name"] 
129                  counts["found-partner-ids"] += 1 
130              else: 
131                  counts["missing-partner-ids"] += 1 
132              if verbose: print "  ", interaction 
133      if verbose: print "Interaction resolution counts:", counts 
134   
135 -def buildInteractionPairs(data): 
136      intPairs = defaultdict(lambda : defaultdict(lambda: False)) 
137      for id in data: 
138          for interaction in data[id]["interaction"]: 
139              if interaction[1] != None: 
140                  intPairs[id][interaction[0]] = True 
141                  intPairs[interaction[0]][id] = True 
142      return intPairs 
143       
144 -def mapNamesToIds(data, normalize=True, verbose=False): 
145      counts = defaultdict(int) 
146      nameToId = defaultdict(list) 
147      for id in sorted(data.keys()): 
148          for name in [data[id]["name"]] + data[id]["synonym"] + data[id]["brand"]: 
149              #assert name not in nameToId, name 
150              if normalize: 
151                  name = normalizeDrugName(name) 
152              if id not in nameToId[name]: 
153                  nameToId[name].append(id) 
154      # count 
155      for name in nameToId: 
156          counts[len(nameToId[name])] += 1 
157          if len(nameToId[name]) > 2: 
158              if verbose: print "Multiple ids:", len(nameToId[name]), name, nameToId[name] 
159      if verbose: print "Name to id:", counts 
160      return nameToId 
161   
162 -def loadDrugBank(filename, preTag="{http://drugbank.ca}", verbose=False): 
163      data = defaultdict(lambda : defaultdict(list)) 
164      print "Loading DrugBank XML" 
165      xml = ETUtils.ETFromObj(filename) 
166      print "Processing DrugBank XML" 
167      root = xml.getroot() 
168      assert root.tag == preTag+"drugs", root.tag 
169      for drug in root.findall(preTag+"drug"): 
170          id = drug.find(preTag+"drugbank-id").text 
171          name = drug.find(preTag+"name").text 
172          if verbose: print id, name 
173          assert id not in data 
174          data[id]["name"] = name 
175          # TODO: Enzymes & targets 
176          # TODO: hydrophobicity 
177          getNestedItems(drug, "synonym", data[id], preTag) 
178          getNestedItems(drug, "brand", data[id], preTag) 
179          getNestedItems(drug, "group", data[id], preTag) 
180          getNestedItems(drug, "category", data[id], preTag, "categories") 
181          interactions = drug.find(preTag+"drug-interactions").findall(preTag+"drug-interaction") 
182          for interaction in interactions: 
183              data[id]["interaction"].append( [interaction.find(preTag+"drug").text, None, interaction.find(preTag+"description").text,] ) 
184      return data 
185   
186 -def prepareDrugBank(drugBankFile): 
187      data = loadDrugBank(drugBankFile) 
188      resolveInteractions(data) 
189      nameToId = mapNamesToIds(data) 
190      return data, nameToId 
191   
192  if __name__=="__main__": 
193  #    drugBankFile = "/home/jari/data/DDIExtraction2011/resources/drugbank.xml" 
194  #    data = loadDrugBank(drugBankFile, verbose=True) 
195  #    nameToId = mapNamesToIds(data, verbose=True) 
196  #    #print nameToId 
197  #    #resolveInteractions(data) 
198      f = DrugFeatureBuilder() 
199      #print f.interactionPairs 
200      print "1:", f.getInteraction("Refludan", "Treprostinil") 
201      print "2:", f.getInteraction("Refludan", "TreprostinilBlahBlah") 
202      print "3:", f.getInteraction("Refludan", "[4-({5-(AMINOCARBONYL)-4-[(3-METHYLPHENYL)AMINO]PYRIMIDIN-2-YL}AMINO)PHENYL]ACETIC ACID") 
203