Approved Research

Using polygenic risk scores, calculated from the largest GWAS to date, as proxy-phenotypes for the brain disorders

Lay summary

The study of brain pathologies associated with aging usually relies on case-control samples, with limited sample size. Population samples such as the UKB, suffer from a healthy bias, which limits studying clinical status. We propose to use polygenic risk scores, calculated from the largest GWAS to date, as proxy-phenotypes for the brain disorders. Such scores should be superior to self-reported family history, collected as part of the UKB. Further meta-analyses of the results obtained on the UKB and smaller clinical samples, should shed light on the brain regions associated with the disorders, but may also result in an improved risk prediction from brain MRI.

Scope extension:

Brain pathologies associated with aging are a major public health issue. These diseases are currently diagnosed too late, due in particular to the difficulty of access to specialised medical centres. Artificial intelligence techniques are a promising way to improve diagnosis.

For the diagnosis of neurodegenerative diseases, deep learning techniques have not yet demonstrated their superiority over traditional machine learning techniques unlike what has been shown in many applications (breast cancer screening, chest radiographs, histology, etc). It can be assumed that this is because the datasets on which these techniques have been trained so far were too small (typically a few hundred patients).

The present project aims to develop a system to assist in the diagnosis of the various pathologies of the aging brain using deep learning techniques on very large databases. In addition, this study will allow a detailed evaluation of the influence of different parameters on the performance and a comparison with traditional statistical and machine learning approaches.

In addition, neurodegenerative diseases (Alzheimer in particularly) are progressive and take decades to become established. Prevention will be key to design efficient treatments by treating the cause of the disease. An additional goal from our team is to run a quantitative and systematic epidemiological study on several large cohorts to show which phenotypes can be considered as risk factors for dementia several years before the appearance of classical symptoms. This study will help to focus on the main risk factors and improve our understanding of the disease.

The study of brain pathologies associated with aging usually relies on case-control samples, with limited sample size. Population samples such as the UKB, suffer from a healthy bias, which limits studying clinical status. We propose to use polygenic risk scores, calculated from the largest GWAS to date, as proxy-phenotypes for the brain disorders. Such scores should be superior to self-reported family history, collected as part of the UKB. Further meta-analyses of the results obtained on the UKB and smaller clinical samples, should shed light on the brain regions associated with the disorders, but may also result in an improved risk prediction from brain MRI.

The hippocampus plays an important role in the development of several age-related disorders including Alzheimer's disease. Incomplete hippocampal inversion (IHI) is an atypical anatomical pattern of the brain that can be evaluated on MRI. It is quite common in the general population and has been linked to several major neurological and psychiatric disorders, mainly epilepsy and schizophrenia. However, the causes of this atypical pattern remain mostly unknown. Moreover, their potential role in the development of age-related neurodegenerative diseases has not been studied. We propose to evaluate the presence of IHI on MRI in UKB participants, to perform a GWAS of IHI and to study their association with other phenotypes. To that purpose, we will visually evaluate IHI in a subsample of the population. We will then train and validate a deep learning method which will be used to evaluate IHI in the whole population. In addition to IHI we will also study anatomical gyrification paterns in the temporal lobe. We propose to manually characterise them before training a deep-learning algorithm that we will apply in the full UKB in order to link gyrification patter with with behaviour and clinical information.

In addition, we propose to perform neuroimaging analyses of risk taking/avoindance behaviour that aim to replicate and extend a previous article (Aydogan et al. 2021). In particular we will report morphometricity (Couvy-Duchesne et al., 2020) and provide a voxel-wise characterisation of the brain regions associated with the phenotype.

In addition, we would like to develop new MRI processing tools that allow extracting brain measurements of brain regions that are not well measured or studied nowadays (e.g. deep nuclei). We will also work on new methods to enhance MRI harmonisation across sites and populations with different demographics. Another project consists in using the sheer number of brain MRI collected in the UKB to pre-train some machine learning algorithms in order to boost prediction accuracy from brain images.