Researches On The Statistical Methods About The Ability Of Gene Classification To Classify Samples

Objective To deeply explore and evaluate the statistic methods applied in the microarray data of human acute leukemia cases. These statistic methods are focus on the gene selection from the large database and applying the capable gene in class discovery of classify samples.Methods Several analysis methods are used to analyze the microarray data of human acute leukemia cases. By comparing the results of statistical analysis of the methods, the suitability of these methods are evaluated and discussed. ResultsAt the 0.005 significance level, we cut the whole data randomly into 7~10 parts, we use stepwise discriminant analysis to select informative genes and 56~80 genes are selected.55 gene variables which have high ability of classification are sieved by percentile analysis and neighborhood analysis. Neighborhood analysis definite the distinguish threshold as r *=0.22. But we chose 893 gene variables which r *≥0.5 to study. There is statistically difference in the means of the two kinds of human acute leukemia( P < 0.05).We discriminated the new samples by the gene variables selected by stepwise discriminant and percentile analysis and neighborhood analysis. Results show that some AML(acute myeloid leukemia) cases are classified wrongly. The probability of misclassification is under 0.20.