目前恶性肿瘤的诊断与亚组分型，主要使用病理组织学方法，作为临床诊断的“金标准”。然而，由于肿瘤组织结构的复杂性和分子异质性，不同类型的肿瘤具有不同的病理组织结构特征和分级标准，实际中经常难以诊断。本课题研究如何在多组学分子层面上对肿瘤进行亚组鉴别，使其与临床诊断、治疗和预后的目标相一致。对肿瘤分子亚组模式识别，需要筛选出不同亚组的生物标志物，进而得到预测模型。然而，对于有些亚组识别问题，实际中可能只有少量的样品存在亚型标签。本研究提出一种新方法，将有标签和无标签的数据混合在一起，使用随机森林（RF）模型和半监督分析算法对数据进行分析，筛选亚组生物标记物和建立预测模型。研究的主要内容：亚组识别的半监督分析模型、半监督分析的多阶段算法、基于样品权重的半监督Boosting算法、半监督分析的变量筛选方法、分子亚组预测模型等。研究拟使用理论推导、模拟实验和结直肠癌亚组实际数据分析相结合的方法进行。

Recently, histopathological method has been regarded as the gold standard for cancer diagnosis and subgroup analysis in clinical practice. However, due to the molecular heterogeneity and complexity of cancer tissues, cancer subtype has different pathological structure characteristics and grading standards, which would increase the difficulty of diagnosis in practice. In regard to these problems, this study aims at identifying the cancer subgroups based on multi-omics molecular information to improve the diagnosis accuracy and therapy and prognosis. To identify molecular subgroups of cancer, it is necessary to select significant biomarkers for each subgroup and then build up the prediction model. However, for some subgourp identification, only a small number of samples have complete and accurate labels in practice. In this study, we propose to combine labeled and unlabeled samples, analyze the data with random forest model and semi-supervised analysis algorithm, select biomarkers and establish prediction model. The main research contents include semi-supervised model for subgroup identification, multi-stage algorithm for semi-supervised analysis, semi-supervised Boosting algorithm based on weight obtained from samples, variable selection method for semi-supervised analysis, molecular subgroup prediction model, and so on. This study intends to combine the method of theoretical derivation, simulation studies and real example analysis with data from colorectal cancer.