A multimodal deep learning model to infer cell-type-specific functional gene networks.

BACKGROUND: Functional gene networks (FGNs) capture functional relationships among genes that vary across tissues and cell types. Construction of cell-type-specific FGNs enables the understanding of cell-type-specific functional gene relationships and insights into genetic mechanisms of human diseases in disease-relevant cell types. However, most existing FGNs were developed without consideration of specific cell types within tissues. RESULTS: In this study, we created a multimodal deep learning model (MDLCN) to predict cell-type-specific FGNs in the human brain by integrating single-nuclei gene expression data with global protein interaction networks. We systematically evaluated the prediction performance of the MDLCN and showed its superior performance compared to two baseline models (boosting tree and convolutional neural network). Based on the predicted cell-type-specific FGNs, we observed that cell-type marker genes had a higher level of hubness than non-marker genes in their corresponding cell type. Furthermore, we showed that risk genes underlying autism and Alzheimer's disease were more strongly connected in disease-relevant cell types, supporting the cellular context of predicted cell-type-specific FGNs. CONCLUSIONS: Our study proposes a powerful deep learning approach (MDLCN) to predict FGNs underlying a diverse set of cell types in human brain. The MDLCN model enhances prediction accuracy of cell-type-specific FGNs compared to single modality convolutional neural network (CNN) and boosting tree models, as shown by higher areas under both receiver operating characteristic (ROC) and precision-recall curves for different levels of independent test datasets. The predicted FGNs also show evidence for the cellular context and distinct topological features (i.e. higher hubness and topological score) of cell-type marker genes. Moreover, we observed stronger modularity among disease-associated risk genes in FGNs of disease-relevant cell types. For example, the strength of connectivity among autism risk genes was stronger in neurons, but risk genes underlying Alzheimer's disease were more connected in microglia.

Deep learning predicts DNA methylation regulatory variants in the human brain and elucidates the genetics of psychiatric disorders.

There is growing evidence for the role of DNA methylation (DNAm) quantitative trait loci (mQTLs) in the genetics of complex traits, including psychiatric disorders. However, due to extensive linkage disequilibrium (LD) of the genome, it is challenging to identify causal genetic variations that drive DNAm levels by population-based genetic association studies. This limits the utility of mQTLs for fine-mapping risk loci underlying psychiatric disorders identified by genome-wide association studies (GWAS). Here we present INTERACT, a deep learning model that integrates convolutional neural networks with transformer, to predict effects of genetic variations on DNAm levels at CpG sites in the human brain. We show that INTERACT-derived DNAm regulatory variants are not confounded by LD, are concentrated in regulatory genomic regions in the human brain, and are convergent with mQTL evidence from genetic association analysis. We further demonstrate that predicted DNAm regulatory variants are enriched for heritability of brain-related traits and improve polygenic risk prediction for schizophrenia across diverse ancestry samples. Finally, we applied predicted DNAm regulatory variants for fine-mapping schizophrenia GWAS risk loci to identify potential novel risk genes. Our study shows the power of a deep learning approach to identify functional regulatory variants that may elucidate the genetic basis of complex traits.

The relationship between DNA methylation in neurotrophic genes and age as evidenced from three independent cohorts: differences by delirium status.

We previously reported the association between DNA methylation (DNAm) of pro-inflammatory cytokine genes and age. In addition, neurotrophic factors are known to be associated with age and neurocognitive disorders. Therefore, we hypothesized that DNAm of neurotrophic genes change with age, especially in delirium patients. DNAm was analyzed using the Illumina HumanMethylation450 or HumanMethylationEPIC BeadChip Kit in 3 independent cohorts: blood from 383 Grady Trauma Project subjects, brain from 21 neurosurgery patients, and blood from 87 inpatients with and without delirium. Both blood and brain samples showed that most of the DNAm of neurotrophic genes were positively correlated with age. Furthermore, DNAm of neurotrophic genes was more positively correlated with age in delirium cases than in non-delirium controls. These findings support our hypothesis that the neurotrophic genes may be epigenetically modulated with age, and this process may be contributing to the pathophysiology of delirium.

Genome-wide DNA methylation investigation of glucocorticoid exposure within buccal samples.

AIM: Glucocorticoids play a major role in regulating the stress response, and an imbalance of glucocorticoids has been implicated in stress-related disorders. Within mouse models, CpGs across the genome have been shown to be differentially methylated in response to glucocorticoid treatment, and using the Infinium 27K array, it was shown that humans given synthetic glucocorticoids had DNA methylation (DNAm) changes in blood. However, further investigation of the extent to which glucocorticoids affect DNAm across a larger proportion of the genome is needed. METHODS: Buccal samples were collected before and after synthetic glucocorticoid treatment in the context of a dental procedure. This included 30 tooth extraction surgery patients who received 10 mg of dexamethasone. Genome-wide DNAm was assessed with the Infinium HumanMethylationEPIC array. RESULTS: Five CpGs showed genome-wide significant DNAm changes that were >10%. These differentially methylated CpGs were in or nearest the following genes: ZNF438, KLHDC10, miR-544 or CRABP1, DPH5, and WDFY2. Using previously published datasets of human blood gene expression changes following dexamethasone exposure, a significant proportion of genes with false-discovery-rate-adjusted significant CpGs were also differentially expressed. A pathway analysis of the genes with false-discovery-rate-adjusted significant CpGs revealed significant enrichment of olfactory transduction, pentose and glucuronate interconversions, ascorbate and aldarate metabolism, and steroid hormone biosynthesis pathways. CONCLUSION: High-dose synthetic glucocorticoid administration in the setting of a dental procedure was significantly associated with DNAm changes within buccal samples. These findings are consistent with prior findings of an influence of glucocorticoids on DNAm in humans.

Genome-wide DNA methylation comparison between live human brain and peripheral tissues within individuals.

Differential DNA methylation in the brain is associated with many psychiatric diseases, but access to brain tissues is essentially limited to postmortem samples. The use of surrogate tissues has become common in identifying methylation changes associated with psychiatric disease. In this study, we determined the extent to which peripheral tissues can be used as surrogates for DNA methylation in the brain. Blood, saliva, buccal, and live brain tissue samples from 27 patients with medically intractable epilepsy undergoing brain resection were collected (age range 5-61 years). Genome-wide methylation was assessed with the Infinium HumanMethylation450 (n = 12) and HumanMethylationEPIC BeadChip arrays (n = 21). For the EPIC methylation data averaged for each CpG across subjects, the saliva-brain correlation (r = 0.90) was higher than that for blood-brain (r = 0.86) and buccal-brain (r = 0.85) comparisons. However, within individual CpGs, blood had the highest proportion of CpGs correlated to brain at nominally significant levels (20.8%), as compared to buccal tissue (17.4%) and saliva (15.1%). For each CpG and each gene, levels of brain-peripheral tissue correlation varied widely. This indicates that to determine the most useful surrogate tissue for representing brain DNA methylation, the patterns specific to the genomic region of interest must be considered. To assist in that objective, we have developed a website, IMAGE-CpG, that allows researchers to interrogate DNA methylation levels and degree of cross-tissue correlation in user-defined locations across the genome.

Epigenetics of Delirium and Aging: Potential Role of DNA Methylation Change on Cytokine Genes in Glia and Blood Along With Aging.

Background: Delirium in elderly patients is common and dangerous. Major risk factors include aging and exogenous insults, such as infection or surgery. In animal models, aging enhances pro-inflammatory cytokine release from microglia in response to exogenous insults. The epigenetic mechanism DNA methylation (DNAm) regulates gene expression and changes with age. Older individuals may have methylation changes that influence the increased cytokine upon insult, but the degree to which aging affects DNAm of cytokine genes is not fully understood. Methods: The relationship between DNAm and aging of pro-inflammatory cytokine genes (TNF-alpha, IL1-beta, IL-6) was investigated using methylation array data in two cohorts. Brain and blood samples were collected from a neurosurgery cohort (NSG) of 21 subjects who underwent brain resection. A second cohort, the Grady Trauma Project (GTP), included blood samples from 265 subjects. Results: In the NSG cohort, a significant negative correlation between age and DNAm in brain was found at a CpG in IL-6. With the GTP dataset, significant negative correlations between age and DNAm were seen at most of the CpGs in TNF-alpha. Also, TNF-Alpha expression increases with age. These GTP DNAm correlations were also nominally significant in NSG blood samples. In neuronal negative NSG brain tissue, a similar negative trend was observed. Conclusions: With aging, a decrease in DNAm of cytokines gene CpGs in glia and blood was seen. As this can affect their expression, additional research is needed to fully elucidate the role of DNAm in aging and how it may influence the pathogenesis of delirium.

Investigation of differential HDAC4 methylation patterns in eating disorders.

The objective of this study was to investigate the relationship between methylation patterns of the histone deacetylase 4 gene and eating disorders in a site previously associated with anorexia nervosa (AN). Women with AN (N=28) or bulimia nervosa (BN) (N=19) were age-matched and sex-matched to controls (N=45). We obtained saliva-derived DNA and use bisulfite pyrosequencing to examine region-specific methylation differences between cases and controls. The region assayed includes 15 CpGs. We found no significant association between the previously implicated CpG and either AN or BN. We found that three CpGs were nominally associated with AN (P=0.02-0.03); the largest difference was a 9% hypermethylation in AN. One CpG was nominally associated with BN (P=0.04), with 4% hypomethylation. None of these results remained significant after correction for multiple testing. We did not replicate previous findings, though through expanded coverage, we identified additional CpGs that were nominally associated with eating disorders.