1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
|
=begin
* Name: lazar.rb
* Description: Lazar model representation
* Author: Andreas Maunz <andreas@maunz.de>, Christoph Helma
* Date: 10/2012
=end
module OpenTox
module Model
class Lazar
include OpenTox
attr_accessor :prediction_dataset
# Check parameters for plausibility
# Prepare lazar object (includes graph mining)
# @param[Array] lazar parameters as strings
# @param[Hash] REST parameters, as input by user
def self.create params
lazar = OpenTox::Model::Lazar.new(File.join($model[:uri],SecureRandom.uuid))
training_dataset = OpenTox::Dataset.new(params[:dataset_uri])
lazar.parameters << {RDF::DC.title => "training_dataset_uri", RDF::OT.paramValue => training_dataset.uri}
if params[:prediction_feature]
resource_not_found_error "No feature '#{params[:prediction_feature]}' in dataset '#{params[:dataset_uri]}'" unless training_dataset.find_feature_uri( params[:prediction_feature] )
else # try to read prediction_feature from dataset
resource_not_found_error "Please provide a prediction_feature parameter" unless training_dataset.features.size == 1
params[:prediction_feature] = training_dataset.features.first.uri
end
lazar[RDF::OT.trainingDataset] = training_dataset.uri
prediction_feature = OpenTox::Feature.new(params[:prediction_feature])
predicted_variable = OpenTox::Feature.find_or_create({RDF::DC.title => "#{prediction_feature.title} prediction", RDF.type => [RDF::OT.Feature, prediction_feature[RDF.type]]})
lazar[RDF::DC.title] = prediction_feature.title
lazar.parameters << {RDF::DC.title => "prediction_feature_uri", RDF::OT.paramValue => prediction_feature.uri}
lazar[RDF::OT.dependentVariables] = prediction_feature.uri
bad_request_error "Unknown prediction_algorithm #{params[:prediction_algorithm]}" if params[:prediction_algorithm] and !OpenTox::Algorithm::Neighbors.respond_to?(params[:prediction_algorithm])
lazar.parameters << {RDF::DC.title => "prediction_algorithm", RDF::OT.paramValue => params[:prediction_algorithm]} if params[:prediction_algorithm]
confidence_feature = OpenTox::Feature.find_or_create({RDF::DC.title => "predicted_confidence", RDF.type => [RDF::OT.Feature, RDF::OT.NumericFeature]})
lazar[RDF::OT.predictedVariables] = [ predicted_variable.uri, confidence_feature.uri ]
case prediction_feature.feature_type
when "classification"
lazar.parameters << {RDF::DC.title => "prediction_algorithm", RDF::OT.paramValue => "weighted_majority_vote"} unless lazar.parameter_value "prediction_algorithm"
lazar[RDF.type] = [RDF::OT.Model, RDF::OTA.ClassificationLazySingleTarget]
when "regression"
lazar.parameters << {RDF::DC.title => "prediction_algorithm", RDF::OT.paramValue => "local_svm_regression"} unless lazar.parameter_value "prediction_algorithm"
lazar[RDF.type] = [RDF::OT.Model, RDF::OTA.RegressionLazySingleTarget]
end
lazar.parameter_value("prediction_algorithm") =~ /majority_vote/ ? lazar.parameters << {RDF::DC.title => "propositionalized", RDF::OT.paramValue => false} : lazar.parameters << {RDF::DC.title => "propositionalized", RDF::OT.paramValue => true}
lazar.parameters << {RDF::DC.title => "min_sim", RDF::OT.paramValue => params[:min_sim].to_f} if params[:min_sim] and params[:min_sim].numeric?
lazar.parameters << {RDF::DC.title => "feature_generation_uri", RDF::OT.paramValue => params[:feature_generation_uri]}
#lazar.parameters["nr_hits"] = params[:nr_hits]
case params["feature_generation_uri"]
when /fminer/
if (params[:nr_hits] == "true")
lazar.parameters << {RDF::DC.title => "feature_calculation_algorithm", RDF::OT.paramValue => "smarts_count"}
else
lazar.parameters << {RDF::DC.title => "feature_calculation_algorithm", RDF::OT.paramValue => "smarts_match"}
end
lazar.parameters << {RDF::DC.title => "similarity_algorithm", RDF::OT.paramValue => "tanimoto"}
lazar.parameters << {RDF::DC.title => "min_sim", RDF::OT.paramValue => 0.3} unless lazar.parameter_value("min_sim")
when /descriptor/
method = params["feature_generation_uri"].split(%r{/}).last.chomp
lazar.parameters << {RDF::DC.title => "feature_calculation_algorithm", RDF::OT.paramValue => method}
lazar.parameters << {RDF::DC.title => "similarity_algorithm", RDF::OT.paramValue => "cosine"}
lazar.parameters << {RDF::DC.title => "min_sim", RDF::OT.paramValue => 0.7} unless lazar.parameter_value("min_sim")
end
bad_request_error "Parameter min_train_performance is not numeric." if params[:min_train_performance] and !params[:min_train_performance].numeric?
lazar.parameters << {RDF::DC.title => "min_train_performance", RDF::OT.paramValue => params[:min_train_performance].to_f} if params[:min_train_performance] and params[:min_train_performance].numeric?
lazar.parameters << {RDF::DC.title => "min_train_performance", RDF::OT.paramValue => 0.1} unless lazar.parameter_value("min_train_performance")
if params[:feature_dataset_uri]
bad_request_error "Feature dataset #{params[:feature_dataset_uri]} does not exist." unless URI.accessible? params[:feature_dataset_uri]
lazar.parameters << {RDF::DC.title => "feature_dataset_uri", RDF::OT.paramValue => params[:feature_dataset_uri]}
lazar[RDF::OT.featureDataset] = params["feature_dataset_uri"]
else
# run feature generation algorithm
feature_dataset_uri = OpenTox::Algorithm::Generic.new(params[:feature_generation_uri]).run(params)
lazar.parameters << {RDF::DC.title => "feature_dataset_uri", RDF::OT.paramValue => feature_dataset_uri}
lazar[RDF::OT.featureDataset] = feature_dataset_uri
end
lazar.put
lazar.uri
end
def predict(params)
@prediction_dataset = OpenTox::Dataset.new
# set instance variables and prediction dataset parameters from parameters
params.each {|k,v|
self.class.class_eval { attr_accessor k.to_sym }
instance_variable_set "@#{k}", v
@prediction_dataset.parameters << {RDF::DC.title => k, RDF::OT.paramValue => v}
}
#["training_compounds", "fingerprints", "training_activities", "training_fingerprints", "query_fingerprint", "neighbors"].each {|k|
["training_compounds", "training_activities", "training_fingerprints", "query_fingerprint", "neighbors"].each {|k|
self.class.class_eval { attr_accessor k.to_sym }
instance_variable_set("@#{k}", [])
}
@prediction_feature = OpenTox::Feature.new @prediction_feature_uri
@predicted_variable = OpenTox::Feature.new @predicted_variable_uri
@predicted_confidence = OpenTox::Feature.new @predicted_confidence_uri
@prediction_dataset.metadata = {
RDF::DC.title => "Lazar prediction for #{@prediction_feature.title}",
RDF::DC.creator => @model_uri,
RDF::OT.hasSource => @model_uri,
RDF::OT.dependentVariables => @prediction_feature_uri,
RDF::OT.predictedVariables => [@predicted_variable_uri,@predicted_confidence_uri]
}
@training_dataset = OpenTox::Dataset.new(@training_dataset_uri)
@feature_dataset = OpenTox::Dataset.new(@feature_dataset_uri)
bad_request_error "No features found in feature dataset #{@feature_dataset.uri}." if @feature_dataset.features.empty?
@similarity_feature = OpenTox::Feature.find_or_create({RDF::DC.title => "#{@similarity_algorithm.capitalize} similarity", RDF.type => [RDF::OT.Feature, RDF::OT.NumericFeature]})
@prediction_dataset.features = [ @predicted_variable, @predicted_confidence, @prediction_feature, @similarity_feature ]
prediction_feature_pos = @training_dataset.features.collect{|f| f.uri}.index @prediction_feature.uri
if @dataset_uri
compounds = OpenTox::Dataset.new(@dataset_uri).compounds
else
compounds = [ OpenTox::Compound.new(@compound_uri) ]
end
@training_fingerprints = @feature_dataset.data_entries
# fill trailing missing values with nil
@training_fingerprints = @training_fingerprints.collect do |values|
values << nil while (values.size < @feature_dataset.features.size)
values
end
@training_compounds = @training_dataset.compounds
feature_names = @feature_dataset.features.collect{ |f| f[RDF::DC.title] }
# one Cdk descriptor may produce several features, e.g., Cdk.WienerNumbers produces Cdk.WienerNumbers.WPATH and Cdk.WienerNumbers.WPOL
# -> strip suffix and use the feature only once
feature_names = feature_names.collect do |f|
if f=~/Cdk/ and f.count(".")==2
f[0..(f.rindex(".")-1)]
else
f
end
end
feature_names.uniq!
query_fingerprints = OpenTox::Algorithm::Descriptor.send( @feature_calculation_algorithm, compounds, feature_names )#.collect{|row| row.collect{|val| val ? val.to_f : 0.0 } }
compounds.each do |compound|
database_activities = @training_dataset.values(compound,@prediction_feature)
if database_activities and !database_activities.empty?
database_activities.each do |database_activity|
@prediction_dataset.add_data_entry compound, @prediction_feature, database_activity
end
next
else
# AM: transform to cosine space
@min_sim = (@min_sim.to_f*2.0-1.0).to_s if @similarity_algorithm =~ /cosine/
@training_activities = @training_dataset.data_entries.collect{|entry|
act = entry[prediction_feature_pos] if entry
@prediction_feature.feature_type=="classification" ? @prediction_feature.value_map.invert[act] : act
}
@query_fingerprint = @feature_dataset.features.collect { |f|
val = query_fingerprints[compound][f.title]
bad_request_error "Can not parse value '#{val}' to numeric" if val and !val.numeric?
val ? val.to_f : 0.0
} # query structure
mtf = OpenTox::Algorithm::Transform::ModelTransformer.new(self)
mtf.transform
prediction = OpenTox::Algorithm::Neighbors.send(@prediction_algorithm,
{ :props => mtf.props,
:activities => mtf.activities,
:sims => mtf.sims,
:value_map => @prediction_feature.feature_type=="classification" ? @prediction_feature.value_map : nil,
:min_train_performance => @min_train_performance
} )
predicted_value = prediction[:prediction]#.to_f
confidence_value = prediction[:confidence]#.to_f
# AM: transform to original space
confidence_value = ((confidence_value+1.0)/2.0).abs if @similarity_algorithm =~ /cosine/
predicted_value = @prediction_feature.value_map[prediction[:prediction].to_i] if @prediction_feature.feature_type == "classification"
end
@prediction_dataset.add_data_entry compound, @predicted_variable, predicted_value
@prediction_dataset.add_data_entry compound, @predicted_confidence, confidence_value
if @compound_uri # add neighbors only for compound predictions
@neighbors.each do |neighbor|
n = neighbor[:compound]
@prediction_feature.feature_type == "classification" ? a = @prediction_feature.value_map[neighbor[:activity]] : a = neighbor[:activity]
@prediction_dataset.add_data_entry n, @prediction_feature, a
@prediction_dataset.add_data_entry n, @similarity_feature, neighbor[:similarity]
#@prediction_dataset << [ n, @prediction_feature.value_map[neighbor[:activity]], nil, nil, neighbor[:similarity] ]
end
end
end # iteration over compounds
@prediction_dataset.put
@prediction_dataset
end
end
end
end
|