Multi-cancer molecular signatures and their interrelationships
Wei-Yi Cheng, Tai-Hsien Ou Yang, Hui Shen, Peter W. Laird, Dimitris Anastassiou, the Cancer Genome Atlas Research Network
备注:
[07.11.2013 v2] Additional authors and acknowledgements for people who contributed to the interpretation of attractor signatures. Summarized table for all 18 signatures. Comments on possible functions
Although cancer is known to be characterized by several unifying biological hallmarks, systems biology has had limited success in identifying molecular signatures present in in all types of cancer. The current availability of rich data sets from many different cancer types provides an opportunity for thorough computational data mining in search of such common patterns. Here we report the identification of 18 "pan-cancer" molecular signatures resulting from analysis of data sets containing values from mRNA expression, microRNA expression, DNA methylation, and protein activity, from twelve different cancer types. The membership of many of these signatures points to particular biological mechanisms related to cancer progression, suggesting that they represent important attributes of cancer in need of being elucidated for potential applications in diagnostic, prognostic and therapeutic products applicable to multiple cancer types.
Patterns in Individual Blood Count Trajectories in the UK Biobank Characterise Disease-Specific Signatures and Anticipate Pan-Cancer Risk
King’s College London(伦敦国王学院);Oxford Immune Algorithmics, Oxford University Innovation and London Institute of Healthcare Engineering(牛津免疫算法公司、牛津大学创新公司和伦敦医疗工程研究所);McGill University(麦吉尔大学);British Society for Research on Ageing(英国衰老研究学会);Algorithmic Dynamics Lab, King’s Institute for Artificial Intelligence(算法动力学实验室,伦敦国王学院人工智能研究所)
Outcome signature genes in breast cancer: is there a unique set?
目的:预测原发性恶性肿瘤组织的转移潜能直接关系到治疗方法的选择。几个微阵列研究产生了其表达谱成功预测存活的基因集(Ramaswamy等2003; Sorlie等2001; van 't Veer等2003)。然而,这些基因组之间的重叠几乎为零。在其他复杂疾病中也观察到这种小的重叠(Lossos et al 2003; Miklos and Maleszka 2004),并且可以解释差异的变量引起了广泛的兴趣。在这种情况下,一个主要的未决问题是,这种差异是否可以仅仅归因于诸如不同技术、不同患者和不同类型的分析等微不足道的原因。结果如下:为了回答这个问题,我们集中在一个单一的乳腺癌数据集,并通过一种方法进行分析,van 't Veer等人使用这种方法来产生一组结果预测基因。我们发现,事实上,所得到的基因集并不是唯一的;它受到用于基因选择的患者子集的强烈影响。同样的分析也可以产生许多同样具有预测性的列表。数据的三个主要特性解释了这种敏感性:(a)许多基因与生存相关;(b)这些相关性之间的差异很小;(c)当对不同的患者子集进行测量时,相关性波动很大。一个可能的生物学解释这些属性进行了讨论。
Multiparametric Deep Learning Tissue Signatures for a Radiological Biomarker of Breast Cancer: Preliminary Results
The Johns Hopkins University School of Medicine(约翰霍普金斯大学医学院);Sidney Kimmel Comprehensive Cancer Center(西德尼·基梅尔综合癌症中心);The Johns Hopkins University(约翰霍普金斯大学);University of Kentucky(肯塔基大学);University of Wisconsin School of Medicine and Public Health(威斯康星大学医学与公共卫生学院)
TCR: A Transformer Based Deep Network for Predicting Cancer Drugs Response
Baidu(百度)
由于肿瘤的异质性,在个性化的基础上预测抗癌药物的临床结果在癌症治疗中具有挑战性。传统的计算工作是通过分子谱来描述药物对个体样本的影响,但由于组学数据的高维性,会出现过拟合,阻碍了模型的临床应用。最近的研究表明,深度学习是一种很有前途的方法,通过学习药物和样本之间的比对模式来建立药物反应模型。然而,现有的研究采用简单的特征融合策略,只考虑作为一个整体表示的药物特征,而忽略了子结构信息,可能发挥重要作用时,比对药物和基因。因此,在本文中,我们提出TCR(Transformer based network for Cancer drug Response)来预测抗癌药物的反应。通过利用注意力机制,TCR能够有效地学习药物原子/亚结构与分子特征之间的相互作用。此外,设计了双损失函数和交叉采样策略,以提高TCR的预测能力。我们表明,TCR优于所有其他方法下的各种数据分割策略的所有评估矩阵(一些显着改善)。大量的实验表明,TCR在独立的体外实验和体内真实患者数据上显示出显着提高的泛化能力。我们的研究强调了TCR的预测能力及其对癌症药物再利用和精确肿瘤治疗的潜在价值。