The authors regret that the original data availability statement of our manuscript was not included in the final submission and this oversight was not identified during proofreading. In the context of open-access publication, it is essential that the correct data availability information is provided. The correct statement is as follows: The data generated and analysed during the current study can be found in the Supplementary Data file with the exception of the single cell RNA sequencing data. The count matrices derived from the raw single cell RNA sequencing data have been uploaded to Gene Expression Omnibus under accession number GSE290973. Due to privacy considerations, the raw sequencing files have not been deposited in the Gene Expression Omnibus but can be obtained from the corresponding author upon reasonable request and completion of a Data Use Agreement. The authors would like to apologise for any inconvenience caused.

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: Many families with clinical early-onset Alzheimer’s disease (EOAD) remain genetically unexplained. A combination of genetic factors is not standardly investigated. In addition to monogenic causes, we evaluated the possible polygenic architecture in a large series of families, to assess if genetic testing of familial EOAD could be expanded. Methods: Thirty-six pedigrees (77 patients) were ascertained from a larger cohort of patients, with relationships determined by genetic data (exome sequencing data and/or SNP arrays). All families included at least one AD patient with symptom onset <70 years. We evaluated segregating rare variants in known dementia-related genes, and other genes or variants if shared by multiple families. APOE was genotyped and duplications in APP were assessed by targeted test or using SNP array data. We computed polygenic risk scores (PRS) compared with a reference population-based dataset, by imputing SNP arrays or exome sequencing data. Results: In eight families, we identified a pathogenic variant, including the genes APP, PSEN1, SORL1, and an unexpected GRN frameshift variant. APOE-ε4 homozygosity was present in eighteen families, showing full segregation with disease in seven families. Eight families harbored a variant of uncertain significance (VUS), of which six included APOE-ε4 homozygous carriers. PRS was not higher in the families combined compared with the population mean (beta 0.05, P = 0.21), with a maximum increase of 0.61 (OR = 1.84) in the GRN family. Subgroup analyses indicated lower PRS in six APP/PSEN1 families compared with the rest (beta −0.22 vs. 0.10; P = 0.009) and lower APOE burden in all eight families with monogenic cause (beta 0.29 vs. 1.15, P = 0.010). Nine families remained without a genetic cause or risk factor identified. Conclusion: Besides monogenic causes, we suspect a polygenic disease architecture in multiple families based on APOE and rare VUS. The risk conveyed by PRS is modest across the studied families. Families without any identified risk factor render suitable candidates for further in-depth genetic evaluation.

Genetic factors play a major role in frontotemporal dementia (FTD). The majority of FTD cannot be genetically explained yet and it is likely that there are still FTD risk loci to be discovered. Common variants have been identified with genome-wide association studies (GWAS), but these studies have not systematically searched for rare variants. To identify rare and new common variant FTD risk loci and provide more insight into the heritability of C9ORF72-related FTD, we performed a GWAS consisting of 354 FTD patients (including and excluding N = 28 pathological repeat carriers) and 4209 control subjects. The Haplotype Reference Consortium was used as reference panel, allowing for the imputation of rare genetic variants. Two rare genetic variants nearby C9ORF72 were strongly associated with FTD in the discovery (rs147211831: OR = 4.8, P = 9.2 × 10⁻⁹, rs117204439: OR = 4.9, P = 6.0 × 10⁻⁹) and replication analysis (P < 1.1 × 10⁻³). These variants also significantly associated with amyotrophic lateral sclerosis in a publicly available dataset. Using haplotype analyses in 1200 individuals, we showed that these variants tag a sub-haplotype of the founder haplotype of the repeat expansion that was previously found to be present in virtually all pathological C9ORF72 G₄C₂ repeat lengths. This new risk haplotype was 10 times more likely to contain a C9ORF72 pathological repeat length compared to founder haplotypes without one of the two risk variants (~22% versus ~2%; P = 7.70 × 10⁻⁵⁸). In haplotypes without a pathologic expansion, the founder risk haplotype had a higher number of repeats (median = 12 repeats) compared to the founder haplotype without the risk variants (median = 8 repeats) (P = 2.05 × 10⁻²⁶⁰). In conclusion, the identified risk haplotype, which is carried by ~4% of all individuals, is a major risk factor for pathological repeat lengths of C9ORF72 G₄C₂. These findings strongly indicate that longer C9ORF72 repeats are unstable and more likely to convert to germline pathological C9ORF72 repeat expansions.

Purpose: To present a fully automatic method to estimate the corneal endothelium parameters from specular microscopy images and to use it to study a one-year follow-up after ultrathin Descemet stripping automated endothelial keratoplasty. Methods: We analyzed 383 post ultrathin Descemet stripping automated endothelial keratoplasty images from 41 eyes acquired with a Topcon SP-1P specular microscope at 1, 3, 6, and 12 months after surgery. The estimated parameters were endothelial cell density (ECD), coefficient of variation (CV), and hexagonality (HEX). Manual segmentation was performed in all images. Results: Our method provided an estimate for ECD, CV, and HEX in 98.4% of the images, whereas Topcon’s software had a success rate of 71.5% for ECD/CV and 30.5% for HEX. For the images with estimates, the percentage error in our method was 2.5% for ECD, 5.7% for CV, and 5.7% for HEX, whereas Topcon’s software provided an error of 7.5% for ECD, 17.5% for CV, and 18.3% for HEX. Our method was significantly better than Topcon’s (P < 0.0001) and was not statistically significantly different from the manual assessments (P > 0.05). At month 12, the subjects presented an average ECD = 1377 ± 483 [cells/mm² ], CV = 26.1 ± 5.7 [%], and HEX = 58.1 ± 7.1 [%]. Conclusions: The proposed method obtains reliable and accurate estimations even in challenging specular images of pathologic corneas. Translational Relevance: CV and HEX, not currently used in the clinic owing to a lack of reliability in automatic methods, are useful biomarkers to analyze the postoperative healing process. Our accurate estimations allow now for their clinical use.

Next-generation sequencing has contributed to our understanding of the genetics of Alzheimer's disease (AD) and has explained a substantial part of the missing heritability of familial AD. We sequenced 19 exomes from 8 Dutch families with a high AD burden and identified EIF2AK3, encoding for protein kinase RNA-like endoplasmic reticulum kinase (PERK), as a candidate gene. Gene-based burden analysis in a Dutch AD exome cohort containing 547 cases and 1070 controls showed a significant association of EIF2AK3 with AD (OR 1.84 [95% CI 1.07–3.17], p-value 0.03), mainly driven by the variant p.R240H. Genotyping of this variant in an additional cohort from the Rotterdam Study showed a trend toward association with AD (p-value 0.1). Immunohistochemical staining with pPERK and peIF2α of 3 EIF2AK3 AD carriers showed an increase in hippocampal neuronal cells expressing these proteins compared with nondemented controls, but no difference was observed in AD noncarriers. This study suggests that rare variants in EIF2AK3 may be associated with disease risk in AD.

In images of the corneal endothelium (CE) acquired by specular microscopy, endothelial cells are commonly only visible in a part of the image due to varying contrast, mainly caused by challenging imaging conditions as a result of a strongly curved endothelium. In order to estimate the morphometric parameters of the corneal endothelium, the analyses need to be restricted to trustworthy regions - the region of interest (ROI) - where individual cells are discernible. We developed an automatic method to find the ROI by Dense U-nets, a densely connected network of convolutional layers. We tested the method on a heterogeneous dataset of 140 images, which contains a large number of blurred, noisy, and/or out of focus images, where the selection of the ROI for automatic biomarker extraction is vital. By using edge images as input, which can be estimated after retraining the same network, Dense U-net detected the trustworthy areas with an accuracy of 98.94% and an area under the ROC curve (AUC) of 0.998, without being affected by the class imbalance (9:1 in our dataset). After applying the estimated ROI to the edge images, the mean absolute percentage error (MAPE) in the estimated endothelial parameters was 0.80% for ECD, 3.60% for CV, and 2.55% for HEX.

Clinical parameters related to the corneal endothelium can only be estimated by segmenting endothelial cell images. Specular microscopy is the current standard technique to image the endothelium, but its low SNR make the segmentation a complicated task. Recently, we proposed a method to segment such images by starting with an oversegmented image and merging the superpixels that constitute a cell. Here, we show how our merging method provides better results than optimizing the segmentation itself. Furthermore, our method can provide accurate results despite the degree of the initial oversegmentation, resulting into a precision and recall of 0.91 for the optimal oversegmentation.

Accumulating evidence suggests that genetic variants in the SORL1 gene are associated with Alzheimer disease (AD), but a strategy to identify which variants are pathogenic is lacking. In a discovery sample of 115 SORL1 variants detected in 1908 Dutch AD cases and controls, we identified the variant characteristics associated with SORL1 variant pathogenicity. Findings were replicated in an independent sample of 103 SORL1 variants detected in 3193 AD cases and controls. In a combined sample of the discovery and replication samples, comprising 181 unique SORL1 variants, we developed a strategy to classify SORL1 variants into five subtypes ranging from pathogenic to benign. We tested this pathogenicity screen in SORL1 variants reported in two independent published studies. SORL1 variant pathogenicity is defined by the Combined Annotation Dependent Depletion (CADD) score and the minor allele frequency (MAF) reported by the Exome Aggregation Consortium (ExAC) database. Variants predicted strongly damaging (CADD score >30), which are extremely rare (ExAC-MAF <1 × 10 '5) increased AD risk by 12-fold (95% CI 4.2-34.3; P=5 × 10 '9). Protein-truncating SORL1 mutations were all unknown to ExAC and occurred exclusively in AD cases. More common SORL1 variants (ExAC-MAF≥1 × 10 '5) were not associated with increased AD risk, even when predicted strongly damaging. Findings were independent of gender and the APOE-I 4 allele. High-risk SORL1 variants were observed in a substantial proportion of the AD cases analyzed (2%). Based on their effect size, we propose to consider high-risk SORL1 variants next to variants in APOE, PSEN1, PSEN2 and APP for personalized risk assessments in clinical practice.