The Geometrical Learning Theory And Its Application In Viruses And Cancers Classification

With the rapid development of molecular biology, the biologicalclassification and evolution have been researched from macroscopicalscopes to microcosmic scopes. The traditional biological taxonomymethod was based on morphologic, however, the environment affectsindividual appearance seriously and brings misguide to the biologicalclassification. The biological classification and evolution methodsresearched from molecular levels are reliable. Additionally, theclassification for cancers sub-types was researched by many methods,which based on morphologic and cytochemistry check, but they werequite similar on this point. The gene expression technology is provided tosolve this problem and applied in diagnosing cancers successfully.In this paper, the algorithm of geometrical learning(GL) is appliedin viruses and cancers classification successfully,1. Firstly, the features of complete genomes sequences of viruses were analyzed, and then all the complete genomes DNA-sequences wereextracted features, and the feature vectors were mapped into the featurespace to be the samples points. Secondly, parts of them were selected tobe the training samples and constructed a graph in the feature space basedon the geometrical learning method. Finally, the rest samples wereclassified based on the constructed graph. The virus complete genomesDNA-sequences of eight classes which provided by NCBI were tested inthis way, and the correct classification rate reached 94%. Compared withthe classification rate of BLAST algorithm, the experimental result of GL-based is highly closed to the result of BLAST. Additionally, theexperimental result of GL-based method has higher recognition rate thanthe result of SVM-based method.2. In this paper, two sub-types(ALL,AML) of human acute leukemiawere classified by GL-based method. Firstly, the informative genes wereselected by "classifying information index" method, and then the twogeometrical combos of classes (ALL, AML) constructed by the GL-basedmethod in the space. Finally the test samples were test by the recognitionalgorithm of GL-based method for cancer subtypes, the accuracy ratereached to 100%.