Benchmarking of deep neural networks for predicting personal gene expression from DNA sequence highlights shortcomings.

Deep learning methods have recently become the state of the art in a variety of regulatory genomic tasks1-6, including the prediction of gene expression from genomic DNA. As such, these methods promise to serve as important tools in interpreting the full spectrum of genetic variation observed in personal genomes. Previous evaluation strategies have assessed their predictions of gene expression across genomic regions; however, systematic benchmarking is lacking to assess their predictions across individuals, which would directly evaluate their utility as personal DNA interpreters. We used paired whole genome sequencing and gene expression from 839 individuals in the ROSMAP study7 to evaluate the ability of current methods to predict gene expression variation across individuals at varied loci. Our approach identifies a limitation of current methods to correctly predict the direction of variant effects. We show that this limitation stems from insufficiently learned sequence motif grammar and suggest new model training strategies to improve performance.

Benchmarking of deep neural networks for predicting personal gene expression from DNA sequence highlights shortcomings.

Deep learning methods have recently become the state-of-the-art in a variety of regulatory genomic tasks1-6 including the prediction of gene expression from genomic DNA. As such, these methods promise to serve as important tools in interpreting the full spectrum of genetic variation observed in personal genomes. Previous evaluation strategies have assessed their predictions of gene expression across genomic regions, however, systematic benchmarking is lacking to assess their predictions across individuals, which would directly evaluates their utility as personal DNA interpreters. We used paired Whole Genome Sequencing and gene expression from 839 individuals in the ROSMAP study7 to evaluate the ability of current methods to predict gene expression variation across individuals at varied loci. Our approach identifies a limitation of current methods to correctly predict the direction of variant effects. We show that this limitation stems from insufficiently learnt sequence motif grammar, and suggest new model training strategies to improve performance.

Aging the brain: multi-region methylation principal component based clock in the context of Alzheimer's disease.

Alzheimer's disease (AD) risk increases exponentially with age and is associated with multiple molecular hallmarks of aging, one of which is epigenetic alterations. Epigenetic age predictors based on 5' cytosine methylation (DNAm), or epigenetic clocks, have previously suggested that epigenetic age acceleration may occur in AD brain tissue. Epigenetic clocks are promising tools for the quantification of biological aging, yet we hypothesize that investigation of brain aging in AD will be assisted by the development of brain-specific epigenetic clocks. Therefore, we generated a novel age predictor termed PCBrainAge that was trained solely in cortical samples. This predictor utilizes a combination of principal components analysis and regularized regression, which reduces technical noise and greatly improves test-retest reliability. To characterize the scope of PCBrainAge's utility, we generated DNAm data from multiple brain regions in a sample from the Religious Orders Study and Rush Memory and Aging Project. PCBrainAge captures meaningful heterogeneity of aging: Its acceleration demonstrates stronger associations with clinical AD dementia, pathologic AD, and APOE ε4 carrier status compared to extant epigenetic age predictors. It further does so across multiple cortical and subcortical regions. Overall, PCBrainAge's increased reliability and specificity makes it a particularly promising tool for investigating heterogeneity in brain aging, as well as epigenetic alterations underlying AD risk and resilience.

Cascading epigenomic analysis for identifying disease genes from the regulatory landscape of GWAS variants.

The majority of genetic variants detected in genome wide association studies (GWAS) exert their effects on phenotypes through gene regulation. Motivated by this observation, we propose a multi-omic integration method that models the cascading effects of genetic variants from epigenome to transcriptome and eventually to the phenome in identifying target genes influenced by risk alleles. This cascading epigenomic analysis for GWAS, which we refer to as CEWAS, comprises two types of models: one for linking cis genetic effects to epigenomic variation and another for linking cis epigenomic variation to gene expression. Applying these models in cascade to GWAS summary statistics generates gene level statistics that reflect genetically-driven epigenomic effects. We show on sixteen brain-related GWAS that CEWAS provides higher gene detection rate than related methods, and finds disease relevant genes and gene sets that point toward less explored biological processes. CEWAS thus presents a novel means for exploring the regulatory landscape of GWAS variants in uncovering disease mechanisms.

Genetics of Gene Expression in the Aging Human Brain Reveal TDP-43 Proteinopathy Pathophysiology.

Here, we perform a genome-wide screen for variants that regulate the expression of gene co-expression modules in the aging human brain; we discover and replicate such variants in the TMEM106B and RBFOX1 loci. The TMEM106B haplotype is known to influence the accumulation of TAR DNA-binding protein 43 kDa (TDP-43) proteinopathy, and the haplotype's large-scale transcriptomic effects include the dysregulation of lysosomal genes and alterations in synaptic gene splicing that are also seen in the pathophysiology of TDP-43 proteinopathy. Further, a variant near GRN, another TDP-43 proteinopathy susceptibility gene, shows concordant effects with the TMEM106B haplotype. Leveraging neuropathology data from the same participants, we also show that TMEM106B and APOE-amyloid-ß effects converge to alter myelination and lysosomal gene expression, which then contributes to TDP-43 accumulation. These results advance our mechanistic understanding of the TMEM106B TDP-43 risk haplotype and uncover a transcriptional program that mediates the converging effects of APOE-amyloid-ß and TMEM106B on TDP-43 aggregation in older adults.

Association of DNA methylation in the brain with age in older persons is confounded by common neuropathologies.

DNA methylation plays a crucial role in the regulation of gene expression, cell differentiation and development. Previous studies have reported age-related alterations of methylation levels in the human brain across the lifespan, but little is known about whether the observed association with age is confounded by common neuropathologies among older persons. Using genome-wide DNA methylation data from 740 postmortem brains, we interrogated 420,132 CpG sites across the genome in a cohort of individuals with ages from 66 to 108 years old, a range of ages at which many neuropathologic indices become quite common. We compared the association of DNA methylation prior to and following adjustment for common neuropathologies using a series of linear regression models. In the simplest model adjusting for technical factors including batch effect and bisulfite conversion rate, we found 8156 CpGs associated with age. The number of CpGs associated with age dropped by more than 10% following adjustment for sex. Notably, after adjusting for common neuropathologies, the total number of CpGs associated with age was reduced by approximately 40%, compared to the sex-adjusted model. These data illustrate that the association of methylation changes in the brain with age is inflated if one does not account for age-related brain pathologies. This article is part of a Directed Issue entitled: Epigenetics dynamics in development and disease.

Morton's foot and pyridoxal 5'-phosphate deficiency: genetically linked traits.

Vitamin B6 is an essential vitamin needed for many chemical reactions in the human body. It exists as several vitamins forms but pyridoxal 5'-phosphate (PLP) is the phosphorylated form needed for transamination, deamination, and decarboxylation. PLP is important in the production of neurotransmitters, acts as a Schiff base and is essential in the metabolism of homocysteine, a toxic amino acid involved in cardiovascular disease, stroke, thrombotic and Alzheimer's disease. This report announces the connection between a deficit of PLP with a genetically linked physical foot form known as the Morton's foot. Morton's foot has been associated with fibromyalgia/myofascial pain syndrome. Another gene mutation methylenetetrahydrofolate reductase (MTHFr) is now being recognized much commonly than previous with chronic fatigue, chronic Lyme diseases and as "the missing link" in other chronic diseases. PLP deficiency also plays a role in impaired glucose tolerance and may play a much bigger role in the obesity, diabetes, fatty liver and metabolic syndrome. Without the Schiff-base of PLP acting as an electron sink, storing electrons and dispensing them in the mitochondria, free radical damage occurs! The recognition that a phenotypical expression (Morton's foot) of a gene resulting in deficiency of an important cofactor enzyme pyridoxal 5'-phosphate will hopefully alert physicians and nutritionist to these phenomena. Supplementation with PLP, L5-MTHF, B12 and trimethylglycine should be used in those patients with hyperhomocysteinemia and/or MTHFR gene mutation.