Identifying gene-gene interactions that are highly associated with Body Mass Index using Quantitative Multifactor Dimensionality Reduction (QMDR)

Genetic data visualization using literature text-based neural networks: Examples associated with myocardial infarction

Data visualization is critical to unraveling hidden information from complex and high-dimensional data. Interpretable visualization methods are critical, especially in the biology and medical fields, however, there are limited effective visualization methods for large genetic data. Current visualization methods are limited to lower-dimensional data and their performance suffers if there is missing data. In this study, we propose a literature-based visualization method to reduce high-dimensional data without compromising the dynamics of the single nucleotide polymorphisms (SNP) and textual interpretability. Our method is innovative because it is shown to (1) preserves both global and local structures of SNP while reducing the dimension of the data using literature text representations, and (2) enables interpretable visualizations using textual information. For performance evaluations, we examined the proposed approach to classify various classification categories including race, myocardial infarction event age groups, and sex using several machine learning models on the literature-derived SNP data. We used visualization approaches to examine clustering of data as well as quantitative performance metrics for the classification of the risk factors examined above. Our method outperformed all popular dimensionality reduction and visualization methods for both classification and visualization, and it is robust against missing and higher-dimensional data. Moreover, we found it feasible to incorporate both genetic and other risk information obtained from literature with our method.

Examining SNP-SNP interactions and risk of clinical outcomes in colorectal cancer using multifactor dimensionality reduction based methods

Background: SNP interactions may explain the variable outcome risk among colorectal cancer patients. Examining SNP interactions is challenging, especially with large datasets. Multifactor Dimensionality Reduction (MDR)-based programs may address this problem.

Objectives: 1) To compare two MDR-based programs for their utility; and 2) to apply these programs to sets of MMP and VEGF-family gene SNPs in order to examine their interactions in relation to colorectal cancer survival outcomes.

Methods: This study applied two data reduction methods, Cox-MDR and GMDR 0.9, to study one to three way SNP interactions. Both programs were run using a 5-fold cross validation step and the top models were verified by permutation testing. Prognostic associations of the SNP interactions were verified using multivariable regression methods. Eight datasets, including SNPs from MMP family genes (n = 201) and seven sets of VEGF-family interaction networks (n = 1,517 SNPs) were examined.

Results: ∼90 million potential interactions were examined. Analyses in the MMP and VEGF gene family datasets found several novel 1- to 3-way SNP interactions. These interactions were able to distinguish between the patients with different outcome risks (regression p-values 0.03–2.2E-09). The strongest association was detected for a 3-way interaction including CHRM3.rs665159_EPN1.rs6509955_PTGER3.rs1327460 variants.

Conclusion: Our work demonstrates the utility of data reduction methods while identifying potential prognostic markers in colorectal cancer.

An epistatic interaction between pre-natal smoke exposure and socioeconomic status has a significant impact on bronchodilator drug response in African American youth with asthma

BACKGROUND

Asthma is one of the leading chronic illnesses among children in the United States. Asthma prevalence is higher among African Americans (11.2%) compared to European Americans (7.7%). Bronchodilator medications are part of the first-line therapy, and the rescue medication, for acute asthma symptoms. Bronchodilator drug response (BDR) varies substantially among different racial/ethnic groups. Asthma prevalence in African Americans is only 3.5% higher than that of European Americans, however, asthma mortality among African Americans is four times that of European Americans; variation in BDR may play an important role in explaining this health disparity. To improve our understanding of disparate health outcomes in complex phenotypes such as BDR, it is important to consider interactions between environmental and biological variables.

RESULTS

We evaluated the impact of pairwise and three-variable interactions between environmental, social, and biological variables on BDR in 233 African American youth with asthma using Visualization of Statistical Epistasis Networks (ViSEN). ViSEN is a non-parametric entropy-based approach able to quantify interaction effects using an information-theory metric known as Information Gain (IG). We performed analyses in the full dataset and in sex-stratified subsets. Our analyses identified several interaction models significantly, and suggestively, associated with BDR. The strongest interaction significantly associated with BDR was a pairwise interaction between pre-natal smoke exposure and socioeconomic status (full dataset IG: 2.78%, p = 0.001; female IG: 7.27%, p = 0.004)). Sex-stratified analyses yielded divergent results for females and males, indicating the presence of sex-specific effects.

CONCLUSIONS

Our study identified novel interaction effects significantly, and suggestively, associated with BDR in African American children with asthma. Notably, we found that all of the interactions identified by ViSEN were "pure" interaction effects, in that they were not the result of strong main effects on BDR, highlighting the complexity of the network of biological and environmental factors impacting this phenotype. Several associations uncovered by ViSEN would not have been detected using regression-based methods, thus emphasizing the importance of employing statistical methods optimized to detect both additive and non-additive interaction effects when studying complex phenotypes such as BDR. The information gained in this study increases our understanding and appreciation of the complex nature of the interactions between environmental and health-related factors that influence BDR and will be invaluable to biomedical researchers designing future studies.

Genetic Variations of Ultraconserved Elements in the Human Genome

Ultraconserved elements (UCEs) are among the most popular DNA markers for phylogenomic analysis. In at least three of five placental mammalian genomes (human, dog, cow, mouse, and rat), 2189 UCEs of at least 200 bp in length that are identical have been identified. Most of these regions have not yet been functionally annotated, and their associations with diseases remain largely unknown. This is an important knowledge gap in human genomics with regard to UCE roles in physiologically critical functions, and by extension, their relevance for shared susceptibilities to common complex diseases across several mammalian organisms in the event of their polymorphic variations. In the present study, we remapped the genomic locations of these UCEs to the latest human genome assembly, and examined them for documented polymorphisms in sequenced human genomes. We identified 29,983 polymorphisms within analyzed UCEs, but revealed that a vast majority exhibits very low minor allele frequencies. Notably, only 112 of the identified polymorphisms are associated with a phenotype in the Ensembl genome browser. Through literature analyses, we confirmed associations of 37 (i.e., out of the 112) polymorphisms within 23 UCEs with 25 diseases and phenotypic traits, including, muscular dystrophies, eye diseases, and cancers (e.g., familial adenomatous polyposis). Most reports of UCE polymorphism-disease associations appeared to be not cognizant that their candidate polymorphisms were actually within UCEs. The present study offers strategic directions and knowledge gaps for future computational and experimental work so as to better understand the thus far intriguing and puzzling role(s) of UCEs in mammalian genomes.

The eMERGE genotype set of 83,717 subjects imputed to ~40 million variants genome wide and association with the herpes zoster medical record phenotype

The Electronic Medical Records and Genomics (eMERGE) network is a network of medical centers with electronic medical records linked to existing biorepository samples for genomic discovery and genomic medicine research. The network sought to unify the genetic results from 78 Illumina and Affymetrix genotype array batches from 12 contributing medical centers for joint association analysis of 83,717 human participants. In this report, we describe the imputation of eMERGE results and methods to create the unified imputed merged set of genome‐wide variant genotype data. We imputed the data using the Michigan Imputation Server, which provides a missing single‐nucleotide variant genotype imputation service using the minimac3 imputation algorithm with the Haplotype Reference Consortium genotype reference set. We describe the quality control and filtering steps used in the generation of this data set and suggest generalizable quality thresholds for imputation and phenotype association studies. To test the merged imputed genotype set, we replicated a previously reported chromosome 6 HLA‐B herpes zoster (shingles) association and discovered a novel zoster‐associated loci in an epigenetic binding site near the terminus of chromosome 3 (3p29).

Collective feature selection to identify crucial epistatic variants

Background

Machine learning methods have gained popularity and practicality in identifying linear and non-linear effects of variants associated with complex disease/traits. Detection of epistatic interactions still remains a challenge due to the large number of features and relatively small sample size as input, thus leading to the so-called "short fat data" problem. The efficiency of machine learning methods can be increased by limiting the number of input features. Thus, it is very important to perform variable selection before searching for epistasis. Many methods have been evaluated and proposed to perform feature selection, but no single method works best in all scenarios. We demonstrate this by conducting two separate simulation analyses to evaluate the proposed collective feature selection approach.

Results

Through our simulation study we propose a collective feature selection approach to select features that are in the "union" of the best performing methods. We explored various parametric, non-parametric, and data mining approaches to perform feature selection. We choose our top performing methods to select the union of the resulting variables based on a user-defined percentage of variants selected from each method to take to downstream analysis. Our simulation analysis shows that non-parametric data mining approaches, such as MDR, may work best under one simulation criteria for the high effect size (penetrance) datasets, while non-parametric methods designed for feature selection, such as Ranger and Gradient boosting, work best under other simulation criteria. Thus, using a collective approach proves to be more beneficial for selecting variables with epistatic effects also in low effect size datasets and different genetic architectures. Following this, we applied our proposed collective feature selection approach to select the top 1% of variables to identify potential interacting variables associated with Body Mass Index (BMI) in ~ 44,000 samples obtained from Geisinger's MyCode Community Health Initiative (on behalf of DiscovEHR collaboration).

Conclusions

In this study, we were able to show that selecting variables using a collective feature selection approach could help in selecting true positive epistatic variables more frequently than applying any single method for feature selection via simulation studies. We were able to demonstrate the effectiveness of collective feature selection along with a comparison of many methods in our simulation analysis. We also applied our method to identify non-linear networks associated with obesity.

Another Round of "Clue" to Uncover the Mystery of Complex Traits

A plethora of genetic association analyses have identified several genetic risk loci. Technological and statistical advancements have now led to the identification of not only common genetic variants, but also low-frequency variants, structural variants, and environmental factors, as well as multi-omics variations that affect the phenotypic variance of complex traits in a population, thus referred to as complex trait architecture. The concept of heritability, or the proportion of phenotypic variance due to genetic inheritance, has been studied for several decades, but its application is mainly in addressing the narrow sense heritability (or additive genetic component) from Genome-Wide Association Studies (GWAS). In this commentary, we reflect on our perspective on the complexity of understanding heritability for human traits in comparison to model organisms, highlighting another round of clues beyond GWAS and an alternative approach, investigating these clues comprehensively to help in elucidating the genetic architecture of complex traits.

Penetrance of Polygenic Obesity Susceptibility Loci across the Body Mass Index Distribution

A growing number of single-nucleotide polymorphisms (SNPs) have been associated with body mass index (BMI) and obesity, but whether the effects of these obesity-susceptibility loci are uniform across the BMI distribution remains unclear. We studied the effects of 37 BMI-associated SNPs in 75,230 adults of European ancestry across BMI percentiles by using conditional quantile regression (CQR) and meta-regression (MR) models. The effects of nine SNPs (24%)-rs1421085 (FTO; p = 8.69 × 10-15), rs6235 (PCSK1; p = 7.11 × 10-6), rs7903146 (TCF7L2; p = 9.60 × 10-6), rs11873305 (MC4R; p = 5.08 × 10-5), rs12617233 (FANCL; p = 5.30 × 10-5), rs11672660 (GIPR; p = 1.64 × 10-4), rs997295 (MAP2K5; p = 3.25 × 10-4), rs6499653 (FTO; p = 6.23 × 10-4), and rs3824755 (NT5C2; p = 7.90 × 10-4)-increased significantly across the sample BMI distribution. We showed that such increases stemmed from unadjusted gene interactions that enhanced the effects of SNPs in persons with a high BMI. When 125 height-associated SNPs were analyzed for comparison, only one (<1%), rs6219 (IGF1, p = 1.80 × 10-4), showed effects that varied significantly across height percentiles. Cumulative gene scores of these SNPs (GS-BMI and GS-height) showed that only GS-BMI had effects that increased significantly across the sample distribution (BMI: p = 7.03 × 10-37; height: p = 0.499). Overall, these findings underscore the importance of gene-gene and gene-environment interactions in shaping the genetic architecture of BMI and advance a method for detecting such interactions by using only the sample outcome distribution.

Mediterranean Diet Adherence and Genetic Background Roles within a Web-Based Nutritional Intervention: The Food4Me Study

Mediterranean Diet (MedDiet) adherence has been proven to produce numerous health benefits. In addition, nutrigenetic studies have explained some individual variations in the response to specific dietary patterns. The present research aimed to explore associations and potential interactions between MedDiet adherence and genetic background throughout the Food4Me web-based nutritional intervention. Dietary, anthropometrical and biochemical data from volunteers of the Food4Me study were collected at baseline and after 6 months. Several genetic variants related to metabolic risk features were also analysed. A Genetic Risk Score (GRS) was derived from risk alleles and a Mediterranean Diet Score (MDS), based on validated food intake data, was estimated. At baseline, there were no interactions between GRS and MDS categories for metabolic traits. Linear mixed model repeated measures analyses showed a significantly greater decrease in total cholesterol in participants with a low GRS after a 6-month period, compared to those with a high GRS. Meanwhile, a high baseline MDS was associated with greater decreases in Body Mass Index (BMI), waist circumference and glucose. There also was a significant interaction between GRS and the MedDiet after the follow-up period. Among subjects with a high GRS, those with a high MDS evidenced a highly significant reduction in total carotenoids, while among those with a low GRS, there was no difference associated with MDS levels. These results suggest that a higher MedDiet adherence induces beneficial effects on metabolic outcomes, which can be affected by the genetic background in some specific markers.

Reverse Pathway Genetic Approach Identifies Epistasis in Autism Spectrum Disorders

Although gene-gene interaction, or epistasis, plays a large role in complex traits in model organisms, genome-wide by genome-wide searches for two-way interaction have limited power in human studies. We thus used knowledge of a biological pathway in order to identify a contribution of epistasis to autism spectrum disorders (ASDs) in humans, a reverse-pathway genetic approach. Based on previous observation of increased ASD symptoms in Mendelian disorders of the Ras/MAPK pathway (RASopathies), we showed that common SNPs in RASopathy genes show enrichment for association signal in GWAS (P = 0.02). We then screened genome-wide for interactors with RASopathy gene SNPs and showed strong enrichment in ASD-affected individuals (P < 2.2 x 10-16), with a number of pairwise interactions meeting genome-wide criteria for significance. Finally, we utilized quantitative measures of ASD symptoms in RASopathy-affected individuals to perform modifier mapping via GWAS. One top region overlapped between these independent approaches, and we showed dysregulation of a gene in this region, GPR141, in a RASopathy neural cell line. We thus used orthogonal approaches to provide strong evidence for a contribution of epistasis to ASDs, confirm a role for the Ras/MAPK pathway in idiopathic ASDs, and to identify a convergent candidate gene that may interact with the Ras/MAPK pathway.

Harnessing publicly available genetic data to prioritize lipid modifying therapeutic targets for prevention of coronary heart disease based on dysglycemic risk

Therapeutic interventions that lower LDL-cholesterol effectively reduce the risk of coronary artery disease (CAD). However, statins, the most widely prescribed LDL-cholesterol lowering drugs, increase diabetes risk. We used genome-wide association study (GWAS) data in the public domain to investigate the relationship of LDL-C and diabetes and identify loci encoding potential drug targets for LDL-cholesterol modification without causing dysglycemia. We obtained summary-level GWAS data for LDL-C from GLGC, glycemic traits from MAGIC, diabetes from DIAGRAM and CAD from CARDIoGRAMplusC4D consortia. Mendelian randomization analyses identified a one standard deviation (SD) increase in LDL-C caused an increased risk of CAD (odds ratio [OR] 1.63 (95 % confidence interval [CI] 1.55, 1.71), which was not influenced by removing SNPs associated with diabetes. LDL-C/CAD-associated SNPs showed consistent effect directions (binomial P = 6.85 × 10⁻⁵). Conversely, a 1-SD increase in LDL-C was causally protective of diabetes (OR 0.86; 95 % CI 0.81, 0.91), however LDL-cholesterol/diabetes-associated SNPs did not show consistent effect directions (binomial P = 0.15). HMGCR, our positive control, associated with LDL-C, CAD and a glycemic composite (derived from GWAS meta-analysis of four glycemic traits and diabetes). In contrast, PCSK9, APOB, LPA, CETP, PLG, NPC1L1 and ALDH2 were identified as “druggable” loci that alter LDL-C and risk of CAD without displaying associations with dysglycemia. In conclusion, LDL-C increases the risk of CAD and the relationship is independent of any association of LDL-C with diabetes. Loci that encode targets of emerging LDL-C lowering drugs do not associate with dysglycemia, and this provides provisional evidence that new LDL-C lowering drugs (such as PCSK9 inhibitors) may not influence risk of diabetes.