论文部分内容阅读
卵巢癌是目前死亡率最高的妇科疾病之一,利用信息学手段挑选特征肿瘤标志物已被广泛用于包括卵巢癌在内的肿瘤分类、诊断研究。但是研究中单纯以提高分类率为指标而忽视敏感性和特异性的均衡,且模型为多变量或者复杂模型,成本过高,不太适合临床应用。为此,提出一种基于“极少”特征标志物的两步预测模型,利用先期提取的多个特征作敏感性和特异性测试,然后构建特征变量的两步预测模型。先用单个变量预测,在一个变量不能得到可靠结果时,才增加另一变量参与模型。实验显示,筛选出的PPE8+LPE4和PPE8+LPC0两对变量组合的敏感性和特异性显著、均衡,变量之间的相关性较小,且分类结果和4个变量的分类结果相当,与9个变量的分类率只差4%～5%。所提出的基于极少特征标志物的两步预测模型结构简单,在保持相同分类效果的前提下大大减少了用于预测的变量,为实际应用提供方便,同时在一定程度上节约了经济成本。
Ovarian cancer is currently one of the highest mortality of gynecological diseases, the selection of informative tumor markers has been widely used in cancer classification, including ovarian cancer, diagnostic studies. However, in the study simply to improve the classification rate as an indicator to ignore the sensitivity and specificity of the balance, and the model is a multivariate or complex model, the cost is too high, not suitable for clinical applications. To this end, a two-step prediction model based on “minimal ” characteristic markers is proposed, which uses the multiple features extracted earlier for sensitivity and specificity testing, and then constructs a two-step prediction model of characteristic variables. Predict with a single variable first, when one variable can not obtain the reliable result, add another variable to participate in the model. The experimental results showed that the sensitivity and specificity of the two PPE8 + LPE4 and PPE8 + LPC0 combinations were significant and balanced, and the correlation between the variables was small. The classification results were similar to those of the four variables The classification rate of a variable is only 4% ~ 5%. The proposed two-step predictive model based on very few feature markers has a simple structure and greatly reduces the variables used for prediction while maintaining the same classification effect, providing convenience for practical applications and saving economic costs to a certain extent.