Integrating multi-omics data into predictive models has the potential to enhance accuracy, which is essential for precision medicine. In this study, we developed interpretable predictive models for multi-omics data by employing neural networks informed by prior biological knowledge, referred to as visible networks. These neural networks offer insights into the decision-making process and can unveil novel perspectives on the underlying biological mechanisms associated with traits and complex diseases. We tested the performance, interpretability and generalizability for inferring smoking status, subject age and LDL levels using genome-wide RNA expression and CpG methylation data from the blood of the BIOS consortium (four population cohorts, Ntotal = 2940). In a cohort-wise cross-validation setting, the consistency of the diagnostic performance and interpretation was assessed. Performance was consistently high for predicting smoking status with an overall mean AUC of 0.95 (95% CI: 0.90-1.00) and interpretation revealed the involvement of well-replicated genes such as AHRR, GPR15 and LRRN3. LDL-level predictions were only generalized in a single cohort with an R2 of 0.07 (95% CI: 0.05-0.08). Age was inferred with a mean error of 5.16 (95% CI: 3.97-6.35) years with the genes COL11A2, AFAP1, OTUD7A, PTPRN2, ADARB2 and CD34 consistently predictive. For both regression tasks, we found that using multi-omics networks improved performance, stability and generalizability compared to interpretable single omic networks. We believe that visible neural networks have great potential for multi-omics analysis; they combine multi-omic data elegantly, are interpretable, and generalize well to data from different cohorts.

Background: The aim of this study was to compare three surgical interventions for correction of sagittal synostosis-frontobiparietal remodeling (FBR), extended strip craniotomy (ESC), and spring-Assisted correction (SAC)-based on three-dimensional (3D) photogrammetry and operation characteristics. Methods: Patients who were born between 1991 and 2019 and diagnosed with nonsyndromic sagittal synostosis who underwent FBR, ESC, or SAC and had at least one postoperative 3D photogrammetry image taken during one of six follow-up appointments until age 6 were considered for this study. Operative characteristics, postoperative complications, reinterventions, and presence of intracranial hypertension were collected. To assess cranial growth, orthogonal cranial slices and 3D photocephalometric measurements were extracted automatically and evaluated from 3D photogrammetry images. Results: A total of 322 postoperative 3D images from 218 patients were included. After correcting for age and sex, no significant differences were observed in 3D photocephalometric measurements. Mean cranial shapes suggested that postoperative growth and shape gradually normalized with higher occipitofrontal head circumference and intracranial volume values compared with normal values, regardless of type of surgery. Flattening of the vertex seems to persist after surgical correction. The authors' cranial 3D mesh processing tool has been made publicly available as a part of this study. Conclusions: The findings suggest that until age 6, there are no significant differences among the FBR, ESC, and SAC in their ability to correct sagittal synostosis with regard to 3D photocephalometric measurements. Therefore, efforts should be made to ensure early diagnosis so that minimally invasive surgery is a viable treatment option. CLINICAL QUESTION/LEVEL OF EVIDENCE: Therapeutic, III.

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.

Three-dimensional (3D) stereophotogrammetry is a novel imaging technique that has gained popularity in the medical field as a reliable, non-invasive, and radiation-free imaging modality. It uses optical sensors to acquire multiple 2D images from different angles which are reconstructed into a 3D digital model of the subject's surface. The technique proved to be especially useful in craniofacial applications, where it serves as a tool to overcome the limitations imposed by conventional imaging modalities and subjective evaluation methods. The capability to acquire high-dimensional data in a quick and safe manner and archive them for retrospective longitudinal analyses, provides the field with a methodology to increase the understanding of the morphological development of the cranium, its growth patterns and the effect of different treatments over time.This review describes the role of 3D stereophotogrammetry in the evaluation of craniosynostosis, including reliability studies, current and potential clinical use cases, and practical challenges. Finally, developments within the research field are analyzed by means of bibliometric networks, depicting prominent research topics, authors, and institutions, to stimulate new ideas and collaborations in the field of craniofacial 3D stereophotogrammetry.We anticipate that utilization of this modality's full potential requires a global effort in terms of collaborations, data sharing, standardization, and harmonization. Such developments can facilitate larger studies and novel deep learning methods that can aid in reaching an objective consensus regarding the most effective treatments for patients with craniosynostosis and other craniofacial anomalies, and to increase our understanding of these complex dysmorphologies and associated phenotypes.

Confounding bias is a crucial problem when applying machine learning to practice, especially in clinical practice. We consider the problem of learning representations independent to multiple biases. In literature, this is mostly solved by purging the bias information from learned representations. We however expect this strategy to harm the diversity of information in the representation, and thus limiting its prospective usage (e.g., interpretation). Therefore, we propose to mitigate the bias while keeping almost all information in the latent representations, which enables us to observe and interpret them as well. To achieve this, we project latent features onto a learned vector direction, and enforce the independence between biases and projected features rather than all learned features. To interpret the mapping between projected features and input data, we propose projection-wise disentangling: a sampling and reconstruction along the learned vector direction. The proposed method was evaluated on the analysis of 3D facial shape and patient characteristics (N = 5011). Experiments showed that this conceptually simple method achieved state-of-the-art fair prediction performance and interpretability, showing its great potential for clinical applications.

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.

Multivariate methods have the potential to better capture complex relationships that may exist between different biological levels. Multiple Factor Analysis (MFA) is one of the most popular methods to obtain factor scores and measures of discrepancy between data sets. However, singular value decomposition in MFA is based on PCA, which is adequate only if the data is normally distributed, linear or stationary. In addition, including strongly correlated variables can overemphasize the contribution of the estimated components. In this work, we introduced a novel method referred as Independent Multifactorial Analysis (ICA-MFA) to derive relevant features from multiscale data. This method is an extended implementation of MFA, where the component value decomposition is based on Independent Component Analysis. In addition, ICA-MFA incorporates a predictive step based on an Independent Component Regression. We evaluated and compared the performance of ICA-MFA with both, the MFA method and traditional univariate analyses, in a simulation study. We showed how ICA-MFA explained up to 10-fold more variance than MFA and univariate methods. We applied the proposed algorithm in a study of 4057 individuals belonging to the population-based Rotterdam Study with available genetic and neuroimaging data, as well as information about executive cognitive functioning. Specifically, we used ICA-MFA to detect relevant genetic features related to structural brain regions, which in turn were involved, in the mechanisms of executive cognitive function. The proposed strategy makes it possible to determine the degree to which the whole set of genetic and/or neuroimaging markers contribute to the variability of the symptomatology jointly, rather than individually. While univariate results and MFA combinations only explained a limited proportion of variance (less than 2%), our method increased the explained variance (10%) and allowed the identification of significant components that maximize the variance explained in the model. The potential application of the ICA-MFA algorithm constitutes an important aspect of integrating multivariate multiscale data, specifically in the field of Neurogenetics.

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.

To study the relation between the microstructure of white matter in the brain and hearing function in older adults we carried out a population-based, cross-sectional study. In 2562 participants of the Rotterdam Study, we conducted diffusion tensor imaging to determine the microstructure of the white-matter tracts. We performed pure-tone audiogram and digit-in-noise tests to quantify hearing acuity. Poorer white-matter microstructure, especially in the association tracts, was related to poorer hearing acuity. After differentiating the separate white-matter tracts in the left and right hemisphere, poorer white-matter microstructure in the right superior longitudinal fasciculus and the right uncinate fasciculus remained significantly associated with worse hearing. These associations did not significantly differ between middle-aged (51–69 years old) and older (70–100 years old) participants. Progressing age was thus not found to be an effect modifier. In a voxel-based analysis no voxels in the white matter were significantly associated with hearing impairment.

Increasing evidence shows that thinner retinal nerve fiber layer (RNFL) and ganglion cell layer (GCL), assessed on optical coherence tomography (OCT), are reflecting global brain atrophy. Yet, little is known on the relation of these layers with specific brain regions. Using voxel-based analysis, we aimed to unravel specific brain regions associated with these retinal layers. We included 2,235 persons (mean age: 67.3 years, 55% women) from the Rotterdam Study (2007–2012) who had gradable retinal OCT images and brain magnetic resonance imaging (MRI) scans, including diffusion tensor (DT) imaging. Thicknesses of peripapillary RNFL and perimacular GCL were measured using an automated segmentation algorithm. Voxel-based morphometry protocols were applied to process DT-MRI data. We investigated the association between retinal layer thickness with voxel-wise gray matter density and white matter microstructure by performing linear regression models. We found that thinner RNFL and GCL were associated with lower gray matter density in the visual cortex, and with lower fractional anisotropy and higher mean diffusivity in white matter tracts that are part of the optic radiation. Furthermore, thinner GCL was associated with lower gray matter density of the thalamus. Thinner RNFL and GCL are associated with gray and white matter changes in the visual pathway suggesting that retinal thinning on OCT may be specifically associated with changes in the visual pathway rather than with changes in the global brain. These findings may serve as a basis for understanding visual symptoms in elderly patients, patients with Alzheimer's disease, or patients with posterior cortical atrophy.

Background: The combination of genetics and imaging has improved their understanding of the brain through studies of aggregate measures obtained from high-resolution structural imaging. Voxel-wise analyses have the potential to provide more detailed information of genetic influences on the brain. Here they report a large-scale study of the heritability of gray matter at voxel resolution (1 × 1 × 1 mm). Methods: Validated voxel-based morphometry (VBM) protocols were applied to process magnetic resonance imaging data of 3,239 unrelated subjects from a population-based study and 491 subjects from two family-based studies. Genome-wide genetic data was used to estimate voxel-wise gray matter heritability of the unrelated subjects and pedigree-structure was used to estimate heritability in families. They subsequently associated two genetic variants, known to be linked with subcortical brain volume, with most heritable voxels to determine if this would enhance their association signals. Results: Voxels significantly heritable in both estimates mapped to subcortical structures, but also voxels in the language areas of the left hemisphere were found significantly heritable. When comparing regional patterns of heritability, family-based estimates were higher than population-based estimates. However, regional consistency of the heritability measures across study designs was high (Pearson's correlation coefficient = 0.73, P = 2.6 × 10⁻¹³). They further show enhancement of the association signal of two previously discovered genetic loci with subcortical volume by using only the most heritable voxels. Conclusion: Gray matter voxel-wise heritability can be reliably estimated with different methods. Combining heritability estimates from multiple studies is feasible to construct reliable heritability maps of gray matter voxels. Hum Brain Mapp 38:2408–2423, 2017.

Background: Multiple sclerosis (MS) affects brain structure and cognitive function and has a heritable component. Over a 100 common genetic risk variants have been identified, but most carriers do not develop MS. For other neurodegenerative diseases, risk variants have effects outside patient populations, but this remains uninvestigated for MS. Objectives: To study the effect of MS-associated genetic variants on brain structure and cognitive function in the general population. Methods: We studied middle-aged and elderly individuals (mean age = 65.7 years) from the population-based Rotterdam Study. We determined 107 MS variants and additionally created a risk score combining all variants. Magnetic resonance imaging (N = 4710) was performed to obtain measures of brain macrostructure, white matter microstructure, and gray matter voxel-based morphometry. A cognitive test battery (N = 7556) was used to test a variety of cognitive domains. Results: The MS risk score was associated with smaller gray matter volume over the whole brain (β_standardized = −0.016; p = 0.044), but region-specific analyses did not survive multiple testing correction. Similarly, no significant associations with brain structure were observed for individual variants. For cognition, rs2283792 was significantly associated with poorer memory (β = −0.064; p = 3.4 × 10⁻⁵). Conclusion: Increased genetic susceptibility to MS may affect brain structure and cognition in persons without disease, pointing to a “hidden burden” of MS.

The volumes of subcortical brain structures are highly heritable, but genetic underpinnings of their shape remain relatively obscure. Here we determine the relative contribution of genetic factors to individual variation in the shape of seven bilateral subcortical structures: the nucleus accumbens, amygdala, caudate, hippocampus, pallidum, putamen and thalamus. In 3,686 unrelated individuals aged between 45 and 98 years, brain magnetic resonance imaging and genotyping was performed. The maximal heritability of shape varies from 32.7 to 53.3% across the subcortical structures. Genetic contributions to shape extend beyond influences on intracranial volume and the gross volume of the respective structure. The regional variance in heritability was related to the reliability of the measurements, but could not be accounted for by technical factors only. These findings could be replicated in an independent sample of 1,040 twins. Differences in genetic contributions within a single region reveal the value of refined brain maps to appreciate the genetic complexity of brain structures.

The neural substrate of genetic risk variants for Alzheimer's disease (AD) remains unknown. We studied their effect on healthy brain morphology to provide insight into disease etiology in the preclinical phase. We included 4071 nondemented, elderly participants of the population-based Rotterdam Study who underwent brain magnetic resonance imaging and genotyping. We performed voxel-based morphometry (VBM) on all gray-matter voxels for 19 previously identified, common AD risk variants. Whole-brain expression data from the Allen Human Brain Atlas was used to examine spatial overlap between VBM association results and expression of genes in AD risk loci regions. Brain regions most significantly associated with AD risk variants were the left postcentral gyrus with ABCA7 (rs4147929, p = 4.45 × 10⁻⁶), right superior frontal gyrus by ZCWPW1 (rs1476679, p = 5.12 × 10⁻⁶), and right postcentral gyrus by APOE (p = 6.91 × 10⁻⁶). Although no individual voxel passed multiple-testing correction, we found significant spatial overlap between the effects of AD risk loci on VBM and the expression of genes (MEF2C, CLU, and SLC24A4) in the Allen Brain Atlas. Results are available online on www.imagene.nl/ADSNPs/. In this single largest imaging genetics data set worldwide, we found that AD risk loci affect cortical gray matter in several brain regions known to be involved in AD, as well as regions that have not been implicated before.

High-throughput technology can now provide rich information on a person's biological makeup and environmental surroundings. Important discoveries have been made by relating these data to various health outcomes in fields such as genomics, proteomics, and medical imaging. However, cross-investigations between several high-throughput technologies remain impractical due to demanding computational requirements (hundreds of years of computing resources) and unsuitability for collaborative settings (terabytes of data to share). Here we introduce the HASE framework that overcomes both of these issues. Our approach dramatically reduces computational time from years to only hours and also requires several gigabytes to be exchanged between collaborators. We implemented a novel meta-analytical method that yields identical power as pooled analyses without the need of sharing individual participant data. The efficiency of the framework is illustrated by associating 9 million genetic variants with 1.5 million brain imaging voxels in three cohorts (total N = 4,034) followed by meta-analysis, on a standard computational infrastructure. These experiments indicate that HASE facilitates high-dimensional association studies enabling large multicenter association studies for future discoveries.

Both normal aging and neurodegenerative diseases such as Alzheimer's disease cause morphological changes of the brain. To better distinguish between normal and abnormal cases, it is necessary to model changes in brain morphology owing to normal aging. To this end, we developed a method for analyzing and visualizing these changes for the entire brain morphology distribution in the general aging population. The method is applied to 1000 subjects from a large population imaging study in the elderly, from which 900 were used to train the model and 100 were used for testing. The results of the 100 test subjects show that the model generalizes to subjects outside the model population. Smooth percentile curves showing the brain morphology changes as a function of age and spatiotemporal atlases derived from the model population are publicly available via an interactive web application at agingbrain.bigr.nl.