br Generate new OGA SVM generation
10. Generate new OGA-SVM generation from combining chromo-somes in k, h, w, e.
11. Replace old generation with the new generation contained all Dexmedetomidine produced from elitism, crossover, and mutation algorithms.
The parameters used in Nested-GA are described in Table 2.
Fig. 4. Crossover mechanism.
Nested-GA parameters description.
Parameter Description Value
N number of Microarray genes (features in OuterGA) 17,814 M number of CpG sites (features in InnerGA) 27,578 Fg reduced features (Microarray genes) by T-test 3000 Fc reduced features (CpG sites) by T-test 10,000 Outer-PSize number of chromosomes in OuterGA population 15 Inner-PSize number of chromosomes in InnerGA population 20 n number of genes in a chromosome for OuterGA 20 m number of genes in a chromosome for InnerGA 50 Outer-maxIter maximum number of iterations for OuterGA 100 Inner-maxIter maximum number of iterations for InnerGA 100 J counter increased with each iteration in InnerGA 1 : Outer-maxIter I counter increased with each iteration in OuterGA 1 : Inner-maxIter Outer-Pc probability of crossover for OuterGA 0.5 Inner-Pc probability of crossover for InnerGA 0.5 Outer-Pm probability of mutation for OuterGA 0.1 Inner-Pm probability of mutation for InnerGA 0.1 Outer-E elitism selected chromosomes for OuterGA 1
Inner-E elitism selected chromosomes for InnerGA 2
Outer-C offspring (chromosomes) produced from crossover for OuterGA 7
Inner-C offspring (chromosomes) produced from crossover for InnerGA 10 Outer-U chromosomes produced from mutation for OuterGA 1
Inner-U chromosomes produced from mutation for InnerGA 2
R number of Nested-GA runs 100 goalF required fitness value used in Nested-GA 95% k number of folds in cross-validation for Nested-GA 5
TC1: j maxIter or the desired accuracy from NN (Neural Network) classifier is achieved Inner-maxIter OR goalF TC2: i maxIter or the desired accuracy from SVM (Support Vector Machine) classifier is achieved Outer-maxIter OR goalF
N feature subsets are produced after repeating Nested-GA N Gene Ontology (GO) (Ashburner et al., 2000) and Kyoto Ency-
times. Unique features are picked from those N subsets, and then clopedia of Genes and Genomes (KEGG) pathway (Kanehisa, Goto,
sorted in descending order based on their cumulative frequencies Sato, Furumichi, & Tanabe, 2011) are the most famous Enrichment
over all the N subsets. The more frequent a feature, the higher its Analysis tools. Gene and gene product features across all species
are represented in GO. KEGG pathway is used for mapping genes
to pathways. There are three categories for the GO terms, which
2.2.5. Feature subsets evaluator
are Biological Process (BP), Cellular Component (CC) and Molecular
Ranked features list L is produced from previous stage. Incre-
mental Feature Selection IFS is applied to produce S feature sets.
Starting with the three top ranked features, the first feature set 3. Results and discussion
is constructed. The remaining features are added one by one in-
crementally to produce new feature sets. So each new set is the This section presents the experimental setup and discusses the
previous set with a new feature added. Finally, S feature sets are different evaluation techniques of the proposed Nested Genetic Al-
constructed where the ith feature set is: si = (f1, f2,..., fi+2) where gorithm (Nested-GA) based on the colon cancer datasets detailed
in Section 2.1 and Table 1. The pipeline depicted in Fig. 1 has been