System for predicting metastasis of cancer by statistical pattern recognition with microarrays.

2004年，我们研究了食管鳞状癌（ESCCs）的基因表达谱与浸润深度和淋巴结累及程度的关系。我们使用高密度寡核苷酸微阵列（Affymetrix）检测了54个escc和11个非癌性食管组织中22115个基因的表达。我们的研究结果表明，侵袭深度不同的escc表达了不同的淋巴结转移相关基因，提示从基因表达谱预测escc淋巴结转移时必须考虑浸润深度的程度。2005年，我们对微阵列预测系统的开发进行了全面的综述。在人类基因组分析中，与要检测的基因相比，我们只有很少的样本。样本数量少导致了预测系统的实际困难。基于对肝细胞癌获得的许多结果，我们研究了克服这些困难的策略。最重要的是，应该解决基因选择问题，而不是设计分类器。特别是，处理由于小样本量而引起的可变性的方法至关重要。这导致了强健的基因选择。此外，我们指出，在开发预测系统时，使用监督学习的方法是正确的。

In 2004, we examined gene expression profiles of esophageal squamous carcinomas (ESCCs) with respect to degree of invasive depth and lymph node involvement. We used high-density oligonucleotide microarrays (Affymetrix) to examine expression of 22115 genes in 54 ESCCs and 11 non-cancerous esophageal tissues. Our results show that ESCCs with distinct invasive depths express different genes linked to lymph node metastasis, suggesting that the degree of invasive depth must be considered when predicting lymph node metastasis of ESCCs from gene expression profiles.In 2005, we studied a comprehensive review for developing a predictive system with microarrays. In human genome analysis, we have few samples, compared with the genes to be examined. The small number of samples causes the practical difficulties in a predictive system. On the base of the many results obtained with hepatocellular carcinoma, we studied our strategy to overcome these difficulties. Most importantly, gene selection should be addressed, rather than classifier design. In particular, the way to cope with the variability due to the small sample size is critical. This leads to the robust gene selection. Moreover, we point out that the approach of using supervised learning is right when developing a predictive system.