From single cells to cognitive phenotypes: development of multimodal foundational AI models with clinical applications

神经退行性疾病与不同尺度的变化相关：从细胞水平的低水平变化到表型水平的高水平变化（例如，脑形态学和认知性能的变化）。捕捉不同尺度的变化，并了解这些变化是如何联系在一起的，是更好地了解不同疾病的机制和预后，并确定个性化治疗的基础。实现这一目标的一种方法是收集健康对照和临床人群的认知，成像和单细胞数据的大规模多模式数据集。然而，这是具有挑战性的，耗时且昂贵的。因此，目前缺乏具有这些数据模态的大规模多模态数据集，相反，主要是大的单峰数据集（与最小或没有重叠的参与者跨数据模式）可用。这个博士项目的目的是，因此，双重的：第一章通过实施和训练新的多模态基础AI模型来开发方法，这些模型可用于利用多个当前可用的大型缩放数据集并学习跨数据模态概括的信息构造和变换（例如，单细胞和认知/行为数据），2）通过将这些模型用于具体临床应用的临床可翻译性（例如，亚型发现，疾病预后和精确医学），特别关注学习生物学变化（例如，单细胞基因表达的变化）是不同认知状态的基础，反之亦然。

Neurodegenerative conditions are associated to changes at different scales: from low level changes at the cell level to high level ones at the phenotype level (e.g., changes in brain morphology and cognitive performance). Capturing changes across different scales, and learning how these changes are connected, is fundamental to better understand the mechanisms and prognosis of different conditions, and to identify individualised treatments.A way to achieve this would be to collect large-scale multimodal datasets with cognitive, imaging and single cell data across both healthy controls and clinical populations. However, this is challenging, time consuming and expensive. As a result, large scale multimodal datasets with these data modalities are currently lacking, and, instead, mainly big unimodal datasets (with minimal to no overlap in participants across data modalities) are available.The aims of this PhD project are, therefore, twofold: 1) methodological development through the implementation and training of novel multimodal foundational AI models that can be used to leverage multiple currently available large scale datasets and learn information constructs and transforms that generalise across data modalities (e.g., single cell and cognitive/behavioural data), 2) Clinical translatability by using these models for concrete clinical applications (e.g., subtype discovery, disease prognosis and precision medicine), with a special interest towards learning the biological changes (e.g., changes in single cell gene expression) underlying different cognitive states, and vice-versa.