Applying deep learning in population genomics is challenging because of computational issues and lack of interpretable models. Here, we propose GenNet, a novel open-source deep learning framework for predicting phenotypes from genetic variants. In this framework, interpretable and memory-efficient neural network architectures are constructed by embedding biologically knowledge from public databases, resulting in neural networks that contain only biologically plausible connections. We applied the framework to seventeen phenotypes and found well-replicated genes such as HERC2 and OCA2 for hair and eye color, and novel genes such as ZNF773 and PCNT for schizophrenia. Additionally, the framework identified ubiquitin mediated proteolysis, endocrine system and viral infectious diseases as most predictive biological pathways for schizophrenia. GenNet is a freely available, end-to-end deep learning framework that allows researchers to develop and use interpretable neural networks to obtain novel insights into the genetic architecture of complex traits and diseases.

Finding automatically multiple lesions in large images is a common problem in medical image analysis. Solving this problem can be challenging if, during optimization, the automated method cannot access information about the location of the lesions nor is given single examples of the lesions. We propose a new weakly supervised detection method using neural networks, that computes attention maps revealing the locations of brain lesions. These attention maps are computed using the last feature maps of a segmentation network optimized only with global image-level labels. The proposed method can generate attention maps at full input resolution without need for interpolation during preprocessing, which allows small lesions to appear in attention maps. For comparison, we modify state-of-the-art methods to compute attention maps for weakly supervised object detection, by using a global regression objective instead of the more conventional classification objective. This regression objective optimizes the number of occurrences of the target object in an image, e.g. the number of brain lesions in a scan, or the number of digits in an image. We study the behavior of the proposed method in MNIST-based detection datasets, and evaluate it for the challenging detection of enlarged perivascular spaces – a type of brain lesion – in a dataset of 2202 3D scans with point-wise annotations in the center of all lesions in four brain regions. In MNIST-based datasets, the proposed method outperforms the other methods. In the brain dataset, the weakly supervised detection methods come close to the human intrarater agreement in each region. The proposed method reaches the best area under the curve in two out of four regions, and has the lowest number of false positive detections in all regions, while its average sensitivity over all regions is similar to that of the other best methods. The proposed method can facilitate epidemiological and clinical studies of enlarged perivascular spaces and help advance research in the etiology of enlarged perivascular spaces and in their relationship with cerebrovascular diseases.

The gap between predicted brain age using magnetic resonance imaging (MRI) and chronological age may serve as a biomarker for early-stage neurodegeneration. However, owing to the lack of large longitudinal studies, it has been challenging to validate this link. We aimed to investigate the utility of such a gap as a risk biomarker for incident dementia using a deep learning approach for predicting brain age based on MRI-derived gray matter (GM). We built a convolutional neural network (CNN) model to predict brain age trained on 3,688 dementia-free participants of the Rotterdam Study (mean age 66 ± 11 y, 55% women). Logistic regressions and Cox proportional hazards were used to assess the association of the age gap with incident dementia, adjusted for age, sex, intracranial volume, GM volume, hippocampal volume, white matter hyperintensities, years of education, and APOE ε4 allele carriership. Additionally, we computed the attention maps, which shows which regions are important for age prediction. Logistic regression and Cox proportional hazard models showed that the age gap was significantly related to incident dementia (odds ratio [OR] = 1.11 and 95% confidence intervals [CI] = 1.05-1.16; hazard ratio [HR] = 1.11, and 95% CI = 1.06-1.15, respectively). Attention maps indicated that GM density around the amygdala and hippocampi primarily drove the age estimation. We showed that the gap between predicted and chronological brain age is a biomarker, complimentary to those that are known, associated with risk of dementia, and could possibly be used for early-stage dementia risk screening.

Advances in technology enable increasing amounts of data collection from individuals for biomedical research. Such technologies, for example, in genetics and medical imaging, have also led to important scientific discoveries about health and disease. The combination of multiple types of high-throughput data for complex analyses, however, has been limited by analytical and logistic resources to handle high-dimensional data sets. In our previous EU Joint Programme–Neurodegenerative Disease Research (JPND) Working Group, called HD-READY, we developed methods that allowed successful combination of omics data with neuroimaging. Still, several issues remained to fully leverage high-dimensional multimodality data. For instance, high-dimensional features, such as voxels and vertices, which are common in neuroimaging, remain difficult to harmonize. In this Full-HD Working Group, we focused on such harmonization of high-dimensional neuroimaging phenotypes in combination with other omics data and how to make the resulting ultra-high-dimensional data easily accessible in neurodegeneration research.

Cognition in adults shows variation due to developmental and degenerative components. A recent genomewide association study identified genetic variants for general cognitive function in 148 independent loci. Here, we aimed to elucidate possible developmental and neurodegenerative pathways underlying these genetic variants by relating them to functional, clinical and neuroimaging outcomes. This study was conducted within the population-based Rotterdam Study (N=11,496, mean age 65.3±9.9 years, 58.0% female). We used lead variants for general cognitive function to construct a polygenic score (PGS), and additionally excluded developmental variants at multiple significance thresholds. A higher PGS was related to more years of education (β=0.29, p=4.3×10 ^-7 ) and a larger intracranial volume (β=0.05, p=7.5×10 ^-4 ). To a smaller extent, the PGS was associated with less cognitive decline (βΔG-factor=0.03, p=1.3×10 ^-3 ), which became non-significant after adjusting for education (p=1.6×10 ^-2 ). No associations were found with daily functioning, dementia, parkinsonism, stroke or microstructural white matter integrity. Excluding developmental variants attenuated nearly all associations. In conclusion, this study suggests that the genetic variants identified for general cognitive function are acting mainly through the developmental pathway of cognition. Therefore, cognition, assessed cross-sectionally, seems to have limited value as a biomarker for neurodegeneration.

Background: It is increasingly recognized that the complex functions of human cognition are not accurately represented by arbitrarily-defined anatomical brain regions. Given the considerable functional specialization within such regions, more fine-grained studies of brain structure could capture such localized associations. However, such analyses/studies in a large community-dwelling population are lacking. Objective: To perform a fine-mapping of cognitive ability to cortical and subcortical grey matter on magnetic resonance imaging (MRI). Methods: In 3,813 stroke-free and non-demented persons from the Rotterdam Study (mean age 69.1 (±8.8) years; 55.8% women) with cognitive assessments and brain MRI, we performed voxel-based morphometry and subcortical shape analysis on global cognition and separate tests that tapped into memory, information processing speed, fine motor speed, and executive function domains. Results: We found that the different cognitive tests significantly associated with grey matter density in differential but also overlapping brain regions, primarily in the left hemisphere. Clusters of significantly associated voxels with global cognition were located within multiple anatomic regions: left amygdala, hippocampus, parietal lobule, superior temporal gyrus, insula and posterior temporal lobe. Subcortical shape analysis revealed associations primarily within the head and tail of the caudate nucleus, putamen, ventral part of the thalamus, and nucleus accumbens, more equally distributed among the left and right hemisphere. Within the caudate nucleus both positive (head) as well as negative (tail) associations were observed with global cognition. Conclusions: In a large population-based sample, we mapped cognitive performance to cortical and subcortical grey matter density using a hypothesis-free approach with high-dimensional neuroimaging. Leveraging the power of our large sample size, we confirmed well-known associations as well as identified novel brain regions related to cognition.