Fixing intronsize filtering step when creating the negative training …

…set. Also added in functionality to save out layers

scripts/portcullis/portcullis/rule_filter.py

@@ -276,9 +276,10 @@ def create_training_sets(args):
 			neg_juncs.to_csv(args.prefix + ".neg_layer_" + str(i) + ".tab", sep='\t')
 
 
-	neg_length_limit = int(L95 * 20)
-	print("Intron size L95 =", L95, "negative set will use junctions with intron size over L95 x 20:", neg_length_limit)
+	neg_length_limit = int(L95 * 8)
+	print("Intron size L95 =", L95, "negative set will use junctions with intron size over L95 x 8:", neg_length_limit, "and with maxmmes < 12")
 	neg_juncs = other_juncs.loc[other_juncs["size"] > neg_length_limit]
+	neg_juncs = neg_juncs.loc[neg_juncs["maxmmes"] < 12]
 	neg_set = pd.concat([neg_set, neg_juncs])
 	if args.genuine:
 		print(str(i+1) + "\t" + calcPerformance(neg_juncs, df).longStr())
@@ -388,4 +389,4 @@ def main():
 
 
 
-if __name__=='__main__': main()
+if __name__=='__main__': main()

src/junction_filter.cc

@@ -244,6 +244,10 @@ void portcullis::JunctionFilter::filter() {
             args.push_back(ruleset + "/selftrain_initial_neg.layer7.json");
 
             args.push_back("--prefix=" + output.string() + ".selftrain.initialset");
+
+	    if (this->saveLayers) {
+		    args.push_back("--save_layers");
+	    }
             args.push_back(junctionFile.string());
 
             char* char_args[50];
@@ -648,6 +652,7 @@ int portcullis::JunctionFilter::main(int argc, char *argv[]) {
     bool no_ml;
     bool saveBad;
     bool save_features;
+    bool save_layers;
     bool exongff;
     bool introngff;
     uint32_t max_length;
@@ -719,6 +724,8 @@ int portcullis::JunctionFilter::main(int argc, char *argv[]) {
             "If you wish to use a custom random forest model to filter the junctions file, rather than self-training on the input dataset use this option to. See manual for more details.")
             ("save_features", po::bool_switch(&save_features)->default_value(false),
             "Use this flag to save features (both for training set and again for all junctions) to disk.")
+            ("save_layers", po::bool_switch(&save_layers)->default_value(false),
+            "Use this flag to save to disk each layer produced when creating the training set.")
             ;
     // Positional option for the input bam file
     po::positional_options_description p;
@@ -768,6 +775,7 @@ int portcullis::JunctionFilter::main(int argc, char *argv[]) {
         }
     }
     filter.setSaveFeatures(save_features);
+    filter.setSaveLayers(save_layers);
     filter.setReferenceFile(referenceFile);
     filter.setThreshold(threshold);
     filter.setSmote(!no_smote);

src/junction_filter.hpp

@@ -97,6 +97,7 @@ namespace portcullis {
         uint16_t threads;
         bool saveBad;
         bool saveFeatures;
+        bool saveLayers;
         bool outputExonGFF;
         bool outputIntronGFF;
         uint32_t maxLength;
@@ -192,14 +193,22 @@ namespace portcullis {
             this->train = train;
         }
 
-        bool isSaveBad() const {
+        bool doSaveBad() const {
             return saveBad;
         }
 
         void setSaveBad(bool saveBad) {
             this->saveBad = saveBad;
         }
 
+        bool doSaveLayers() const {
+            return saveLayers;
+        }
+
+        void setSaveLayers(bool saveLayers) {
+            this->saveLayers = saveLayers;
+        }
+
         bool isOutputExonGFF() const {
             return outputExonGFF;
         }

src/portcullis.cc

@@ -185,6 +185,8 @@ int mainFull(int argc, char *argv[]) {
     uint32_t max_length;
     uint32_t mincov;
     string canonical;
+    bool save_layers;
+    bool save_features;
     bool balanced;
     bool verbose;
     bool help;
@@ -258,6 +260,10 @@ int mainFull(int argc, char *argv[]) {
     hidden_options.add_options()
             ("bam-files", po::value< std::vector<path> >(&bamFiles), "Path to the BAM files to process.")
             ("genome-file", po::value<path>(&genomeFile), "Path to the genome file to process.")
+	    ("save_features", po::bool_switch(&save_features)->default_value(false),
+              "Use this flag to save features (both for training set and again for all junctions) to disk.")
+            ("save_layers", po::bool_switch(&save_layers)->default_value(false),
+              "Use this flag to save to disk each layer produced when creating the training set.")
             ;
     // Positional option for the input bam file
     po::positional_options_description p;
@@ -359,6 +365,8 @@ int mainFull(int argc, char *argv[]) {
     filter.setOutputExonGFF(exongff);
     filter.setOutputIntronGFF(introngff);
     filter.setSaveBad(saveBad);
+    filter.setSaveLayers(save_layers);
+    filter.setSaveFeatures(save_features);
     filter.filter();
 
     // *********** BAM filter *********