Discovering early biomarkers of Alzheimer's disease using genetic and physics-informed networks

该项目的目的：获得阿尔茨海默病衍生的皮层扩散和pet的新型生物标志物，利用这些来约束疾病进展的物理信息模型，通过精确定位到艾伦脑图谱来与区域基因表达进行比较。摘要：由于人类大脑的复杂组织结构和高度变异，检测神经退行性疾病的早期生物标志物是一个极具挑战性的问题。据认为，大约50%的痴呆症患者在早期阶段没有得到诊断。这限制了治疗选择，并对临床试验的患者筛选提出了重大挑战。最近的研究表明，皮质微观结构的测量可能是早期神经变性的有效、非侵入性标志物[1,2]。然而，到目前为止，这些测量都被报道为整个大脑的平均汇总测量，而众所周知，细胞组织在整个大脑皮层中存在显著差异，痴呆症的表现在个体之间是不同的。与此同时，最近在小鼠模型中的工作表明，tau病理在大脑中的进展受到神经元连接的极大限制，并且与疾病进展的简单模型的偏差可以通过基因表达[3]很好地解释。类似的启发模型，在人类身上训练，已经使用来自阿尔茨海默病神经成像倡议（ADNI）开放数据集[4]的正电子发射断层扫描（PET）数据进行约束。然而，到目前为止，这些都局限于大脑组织的全球平均模型，而没有考虑个体差异。因此，该项目的目标将是建立个体人类大脑微观结构组织的精确模型[5-10]，并使用这些模型来约束几何深度学习[7,11,12]和生物物理信息神经网络[4,13,14]阿尔茨海默病进展模型。研究结果将与基因表达进行比较，以了解疾病的机制，并改善早期诊断。

Aims of the Project:To derive novel biomarkers of Alzheimer's disease derived cortical diffusion and PETTo use these to constrain physics-informed models of disease progressionTo compare against regional gene-expression through precision mapping to the Allen Brain AtlasSummary:Detecting early biomarkers of neurodegeneration is a highly challenging problem due to the complex organisational structure and high degree of variation of the human brain. Approximately 50% of dementia sufferers are thought to go undiagnosed in early stages. This limits treatment options and presents significant challenges for patient screening for clinical trials.Recent studies have indicated that measures of cortical microstructure may present effective, non-invasive markers of early neurodegeneration [1,2]. However, so far these measures have been reported as summary measures averaged across the brain, when it is well known that cellular organisation varies significantly across the cortex, and that the presentation of dementia varies across individuals.At the same time, recent work in mouse models has shown that the progression of tau pathology through the brain is extremely well constrained by neuronal connectivity, and that deviations from simple models of disease progression can be well explained by gene expression [3]. Similarly inspired models, trained on humans, have been constrained using positron emission tomography (PET) data from the Alzheimer's Disease Neuroimaging Iniative (ADNI) open dataset [4]. However, thus far these have been limited to global average models of brain organisation, not considering individual variability.The goal of this project will therefore be to build precision models of the microstructural organisation of individual human brains [5-10], and to use these to constrain geometric deep learning [7, 11, 12] and biophysically-informed neural networks [4,13,14] models of Alzheimer's disease progression. Findings would be compared against gene expression, to inform mechanistic understanding of the disease, and improve early diagnosis.