Establishment of cDNA array-based predictive diagnostics of cancer for personalized medicine

本项目旨在建立一种基于基因芯片的癌症恶性特征预测方法，为患者的个性化治疗提供依据。我们收集了北海道地区31家医院手术修复的冷冻癌组织标本，并用基因芯片分析了其基因表达谱。通过使用一种新的计算机算法，有效地选择与癌症恶性特征显著相关的特征(基因表达)，我们探索了基因表达的功能意义，并建立了一种有效的个性化方法。1)我们收集了5800多例癌症样本和相关的临床病理数据。2)对其中的600例进行了cDNA阵列分析。3)提出了一种以k近邻为评价函数的特征子集选择方法，一种结合概率神经网络的集成机器学习方法。4)应用该方法，我们可以有效地预测结直肠癌、肺癌和食道癌的淋巴转移，胰腺癌患者的短期生存，胃癌患者的长期生存，以及胆道癌的神经周围侵犯，乳腺癌和肝癌的组织学恶性。这些已在学术会议上报道，并在国际期刊上发表。

This project aimed to establish a predictive method of cDNA array-based prediction of malignant characters of cancers for personalized treatment for patients. We collected frozen cancer tissue samples surgically reseated in 31 hospitals in Hokkaido prefecture and analyzed the mRNA expression profiles with cDNA array analysis. By using of a novel computer algorithm which efficiently select features (gene expressions) that were significantly correlated with malignant characters of cancers, we explored functional significance of the gene expressions and established an effective method of personalization.1)We could collect more than 5,800 cases of cancer samples and the pertinent clinicopathological data.2)Among them, 600 cases were analyzed with cDNA array.3)We developed a new method of feature subset selection utilizing k-nearest neighbor as an assessment function, a method of ensemble machine learning combining probabilistic neural networks.4)By applying this method, we could effectively predict lymph node metastasis of colorectal cancers, lung cancers and esophageal cancers, short term survival of patients with pancreatic cancers, long-term survival of gastric caner patients, and perineural invasion of bile-duct cancers, histological malignancy of breast cancers and liver cancers. These have been reported in academic meetings and published in international journals.