基于遍历基因组合的特征基因选取方法

摘要

Feature gene selection is a hot issue. Under the consumption that cancer is caused by one or some genes heteromorphosis, this article starts research from two genes combination, using Logistic regression method on all possible combinations with 2 genes with prediction accuracy and AIC as the evaluation criteria. Based on the evaluation of all stimulation results, we obtained the best gene combination(X55187, D14812). Meanwhile, we tested and verified the stability of this best combination (X55187, D14812) with leave one out cross validation. At last, we analyzed and compared the frequencies of 640 pairs gene combination which prediction accuracy was more than 90%with former studies. The result shows that prediction accuracy is not high along with the higher frequency gene combination.%特征基因的选取是非常热门的问题，在癌症是由某个或者某几个基因共同相互作用引起变异的假设下，从最简单的2个基因组合进行研究，遍历所有可能的基因组合，运用Logistic回归分类器，以预测精度和AIC准则为评价标准，对所有的模拟结果进行评价，得到最优基因组合（X55187，D14812）。同时运用交叉留一检验，验证了此基因组合建立模型的稳定性。最后又对预测精度大于90%的640对基因组合进行频数分析，并与已有文献进行比较，得到出现频率高的基因组合，预测精度并不一定高的结论。