Multi-class Tumor Subtype Recognition Based On Neural Networks

Building an effective classification model based on gene expression profiles is of great importance in tumor diagnosing and its clinical therapy. To the classification problem of tumor subtype, finding a group of gene subsets that are relevant to tumor subtype is a crucial task. In this work, from the view of information and system science, we were committed to exploring the problem of ALL subtype discrimination and feature selection using methodology of AI and computing technology. The Acute Lymphoblastic Leukemia Tumor dataset(ALL) was taken as a case and the research results are achieved as follows.Firstly, We adopt a measure as weighted Relief distance as a criterion of screening predictive genes for ALL subtype classification. Considering that ALL has seven subtypes, we proposed a improved measure algorithm known as Relief_F as the criterion of gene sorting and selection.Secondly, In view of the research of ALL subtype forecast model, we analyzed a classified tool's application in acute lymphatic leukemia tumor subtype recognition based on neural network. We introduced the characteristic of the neural network and their application in the pattern classification, then we designed a BP network model which has three hidden layer, and the nodes of the hidden layer were changed along with the number of informative genes. Then we selected a subset has 169 genes that could achieve 100% accuracy by the BP network model after leave-one-out cross validation (LOOCV) and independent test (IT).Lastly, we analyzed the filtration of redundant genes in ALL and proposed a new method. We adopt K-means algorithm to analyzed clustering ability of the informative genes and remove its redundant genes. Then we compared with Eng-Juh Yeoh's experimental results, we select an informative gene subset has less 38 genes and we apply artificial neural networks (ANN) to evaluating the classification ability of the selected gene subset. Experiments proved that proposed method is very effective in multi-class tumor subtypes recognition.