Personalising the treatment of brain tumours through advanced graphical modelling

脑癌是英国第九大常见癌症，是导致死亡和残疾的主要原因，至今仍对治疗有顽固的抵抗力。尽管进行了大量的研究，但在过去的30年里，存活率基本上没有变化。胶质母细胞瘤是一种常见的快速生长的脑肿瘤，在确诊后，只有二十分之一的患者有望存活五年，而且生活质量往往严重受损。近几十年来，其他类型的癌症在治疗效果上有了显著的改善：为什么大脑不同？一个可能的答案是脑肿瘤异常复杂。导致不受控制的、无组织的细胞分裂的疾病机制——所有癌症的标志——似乎是多种多样的，并且在每个病人之间差异很大。这里的治疗如果要有效，就必须密切个性化，然而这种多样性使得确定哪种治疗方法非常困难。我们陷入了两难境地：为了确定对每个病人来说什么是最好的，我们需要通过对许多病人的研究来指导，然而，个体的结果是如此多样化，以至于一个大群体的平均值几乎不能告诉我们每个人的情况。我们怎样才能摆脱困境？与所有人类多样性一样，要理解这种生物多样性，我们必须首先对其进行足够详细的描述，使每个个体都是独特的和可识别的。这种丰富的描述或表型可以被认为是每个个体独特的特征“指纹”模式，但可以与其他人进行比较。能够理解复杂模式的机器——比如正在彻底改变世界的人工智能系统——可以被部署到为合适的病人确定合适的治疗方案的任务中，帮助提供既个性化又基于可靠证据的护理。至关重要的是，在处理脑癌的所有NHS医院的常规护理中，已经收集了丰富的详细信息——脑部扫描、组织检查、基因分析——从中可以提取出这种模式，使我们能够采用这种方法，而不会干扰既定的护理途径。但是，目前还没有一个框架可以在整个NHS范围内实现这一目标：我们的任务是建立它的基础，确定它的可行性，并对其应用进行试点。我们的任务分为两部分。首先，我们必须创建复杂的计算机算法，使从脑部扫描、组织样本和遗传学中提取的特征肿瘤模式有意义，并将它们彼此置于有意义的关系中，以便建立它们与患者病程和治疗的联系。其次，我们必须使用来自过去和未来的许多不同患者的大量数据来评估算法的准确性，以验证发现并确保在不同患者群体和医院站点之间的通用性。我们提出的方法的核心是将生物模式理解为网络。打个比方，一个细胞从正常到癌变的过程可以被看作是伦敦地铁的一段旅程，在这里，恶性肿瘤的最终目的地是通过一系列定义特征路径的基因站点到达的。捕获所有可能路径的集合——一张“地下肿瘤”的地图——然后为我们提供了一种理解发生了什么，以及不同患者的疾病过程如何变化的方法。成功将提供一个框架，使个体能够更好地预测患者的结果，更准确、个性化的治疗处方，以及阐明未来治疗创新所依赖的疾病机制的有力手段。

Cancer of the brain - the 9th commonest form in the UK - is a major cause of death and disability that remains stubbornly resistant to treatment. Despite intense research, survival has remained essentially unchanged over the past thirty years. Only one in twenty patients with glioblastoma - a common, fast-growing brain tumour - can expect to be alive five years after diagnosis, and often with severely impaired quality of life. Other kinds of cancer have seen striking improvements in treatment outcomes over recent decades: why is the brain different? One possible answer is the unusual complexity of brain tumours. The disease mechanisms that cause the uncontrolled, disorganised cell division - the hallmark of all cancer - appear to be both many and diverse, and vary greatly from one patient to another. Treatment here must be closely personalised if it is to be effective, yet the diversity makes determining what treatment works very difficult. We are caught in a Catch 22 situation: to determine what is best for each patient we need to be guided by the study of many patients together, yet individual outcomes are so diverse that the average of a large group tells us little about each individual. How do we break out of this? As with all human diversity, to understand this biological diversity we must first describe it in sufficient detail to render each individual distinct and recognizable. Such a rich description or phenotype can be thought of as a characteristic "fingerprint" pattern unique to each individual yet enabling comparison with others. Machines able to make sense of complex patterns - such as the artificial intelligence systems now revolutionising the world - can then be deployed to the task of identifying the right treatment for the right patient, helping deliver care that is both personalised and founded on robust evidence. Crucially, there is already a wealth of detailed information collected during routine care across all NHS hospitals dealing with brain cancer - brain scans, tissue examinations, genetic analyses - from which such patterns may be extracted, allowing us to adopt this approach without disturbing established pathways of care. But no framework to deliver it across the NHS currently exists: our task is to create its foundations, establish its feasibility, and pilot its application. Our task has two parts. First, we must create complex computer algorithms that make sense of characteristic tumour patterns extracted from brain scans, tissue samples, and genetics, placing them into meaningful relation to each other so that their connection with the patient's disease course and treatment can be established. Second, we must assess the accuracy of the algorithms with copious data from many, diverse patients, from both the past and the future, to validate findings and ensure generalisability across both different patient groups and hospital sites. At the heart of our proposed approach lies the idea of making sense of biological patterns as networks. By way of analogy, the path of a cell from normal to cancerous may be seen as a journey through the London Underground, where the final destination of malignancy is reached via a set of genetic stops defining a characteristic path. Capturing the set of all possible paths - a map of the "tumour underground" - then gives us a means of understanding what is going on, and how the disease process varies from one patient to another. Success will provide a framework that will enable better individual prediction of patient outcomes, more accurate, personalised prescription of treatments, and a powerful means of illuminating disease mechanisms on which future treatment innovation depends.