Neural networks for modeling gene-gene interactions in association studies

Background

The monogenic disease osteogenesis imperfecta (OI) is due to single mutations in either of the collagen genes ColA1 or ColA2, but within the same family a given mutation is accompanied by a wide range of disease severity. Although this phenotypic variability implies the existence of modifier gene variants, genome wide scanning of DNA from OI patients has not been reported. Promising genome wide marker-independent physical methods for identifying disease-related loci have lacked robustness for widespread applicability. Therefore we sought to improve these methods and demonstrate their performance to identify known and novel loci relevant to OI.

Results

We have improved methods for enriching regions of identity-by-descent (IBD) shared between related, afflicted individuals. The extent of enrichment exceeds 10- to 50-fold for some loci. The efficiency of the new process is shown by confirmation of the identification of the Col1A2 locus in osteogenesis imperfecta patients from Amish families. Moreover the analysis revealed additional candidate linkage loci that may harbour modifier genes for OI; a locus on chromosome 1q includes COX-2, a gene implicated in osteogenesis.

Conclusion

Technology for physical enrichment of IBD loci is now robust and applicable for finding genes for monogenic diseases and genes for complex diseases. The data support the further investigation of genetic loci other than collagen gene loci to identify genes affecting the clinical expression of osteogenesis imperfecta. The discrimination of IBD mapping will be enhanced when the IBD enrichment procedure is coupled with deep resequencing.     

Early Life Factors and Inter-Country Heterogeneity in BMI Growth Trajectories of European Children: The IDEFICS Study

Background

Starting from birth, this explorative study aimed to investigate between-country differences in body mass index (BMI) trajectories and whether early life factors explain these differences.

Methods

The sample included 7,644 children from seven European countries (Belgium, Cyprus, Germany, Hungary, Italy, Spain, Sweden) participating in the multi-centre IDEFICS study. Information on early life factors and in total 53,409 repeated measurements of height and weight from 0 to <12 years of age were collected during the baseline (2007/2008) and follow-up examination (2009/2010) supplemented by records of routine child health visits. Country-specific BMI growth curves were estimated using fractional polynomial mixed effects models. Several covariates focussing on early life factors were added to the models to investigate their role in the between-countries differences.

Results

Large between-country differences were observed with Italian children showing significantly higher mean BMI values at all ages ≥ 3 years compared to the other countries. For instance, at age 11 years mean BMI values in Italian boys and girls were 22.3 [21.9;22.8; 99% confidence interval] and 22.0 [21.5;22.4], respectively, compared to a range of 18.4 [18.1;18.8] to 20.3 [19.8;20.7] in boys and 18.2 [17.8;18.6] to 20.3 [19.8;20.7] in girls in the other countries. After adjustment for early life factors, differences between country-specific BMI curves became smaller. Maternal BMI was the factor being most strongly associated with BMI growth (p<0.01 in all countries) with associations increasing during childhood. Gestational weight gain (GWG) was weakly associated with BMI at birth in all countries. In some countries, positive associations between BMI growth and children not being breastfed, mothers’ smoking during pregnancy and low educational level of parents were found.

Conclusion

Early life factors seem to explain only some of the inter-country variation in growth. Maternal BMI showed the strongest association with children’s BMI growth.     

Artificial neural networks modeling gene-environment interaction

ABSTRACT: BACKGROUND: Gene-environment interactions play an important role in the etiological pathway of complex diseases. An appropriate statistical method for handling a wide variety of complex situations involving interactions between variables is still lacking, especially when continuous variables are involved. The aim of this paper is to explore the ability of neural networks to model different structures of gene-environment interactions. A simulation study is set up to compare neural networks with standard logistic regression models. Eight different structures of gene-environment interactions are investigated. These structures are characterized by penetrance functions that are based on sigmoid functions or on combinations of linear and non-linear effects of a continuous environmental factor and a genetic factor with main effect or with a masking effect only. RESULTS: In our simulation study, neural networks are more successful in modeling gene-environment interactions than logistic regression models. This outperfomance is especially pronounced when modeling sigmoid penetrance functions, when distinguishing between linear and nonlinear components, and when modeling masking effects of the genetic factor. CONCLUSION: Our study shows that neural networks are a promising approach for analyzing gene-environment interactions. Especially, if no prior knowledge of the correct nature of the relationship between co-variables and response variable is present, neural networks provide a valuable alternative to regression methods that are limited to the analysis of linearly separable data.     

Testing for association in the presence of population stratification: a simulation study comparing the S-TDT, STRAT and the GC

A novel approach for association testing in the presence of population stratification has been introduced by Pritchard et al. (2000a) and Pritchard et al. (2000b). The structured association approach is a two-tiered procedure that first estimates the population structure and then tests the null hypothesis H0: 'no association within subpopulations' in the second step. A power comparison of the stratified test for association (STRAT) (Pritchard et al., 2000b) and the Transmission-Disequilibrium-Test (TDT) (Spielman and Ewens, 1993a) in a simulation framework showed superiority of STRAT if allele frequencies or associations between allele and disease differ strongly in subpopulations. In more homogeneous situations, the TDT had greater power than STRAT. However, the TDT, based on family trios,that uses population controls, needs 50% more genotyping compared to STRAT. The Sib-Transmission-Disequilibrium-Test (S-TDT) needs the same amount of genotyping since it relays in its minimal configuration on pairs of siblings. This raises the question how the S-TDT (Spielman and Ewens, 1998a) performs compared to the population based methods STRAT and Genomic Controls (GC). In this paper, we present a simulation study accounting for two different models of population stratification in different settings of allele frequencies and under different risk models. The results showed that under a discrete as well as under an admixed population model, STRAT strongly outperformed the S-TDT and the GC when different alleles were associated in different subpopulations. In contrast, the S-TDT had greater power than STRAT when the same allele was associated in both subpopulations. Here, the GC was sometimes even more powerful than the S-TDT, depending on the population model and the allele frequency differences. A general recommendation for the use of one of the tests can therefore not be given.     

Hair Minerals and Metabolic Health in Belgian Elementary School Girls

Literature has repeatedly shown a relationship between hair minerals and metabolic health, although studies in children are currently lacking. This study aims to investigate hair levels of calcium (Ca), copper (Cu), magnesium (Mg), iron (Fe), phosphorus (P), and zinc (Zn) and their association with (1) overweight/obesity and (2) metabolic health in Flemish elementary school girls between 5 and 10 years old. Two hundred eighteen girls participated in this study as part of the baseline ChiBS project. Children were subjected to physical examinations, blood and hair sampling. Hair minerals were quantitatively determined via inductively coupled plasma–mass spectrometry after microwave-assisted acid digestion. Body mass index (BMI) and body fat percentage (BF%) were studied as anthropometric parameters, and a metabolic score (including systolic and diastolic blood pressure, insulin resistance and non-high-density lipoprotein (non-HDL) cholesterol as parameters) was calculated, with higher scores indicating a more unhealthy metabolic profile. Hair Ca, Ca/Mg, and Ca/P positively correlated with the anthropometric parameters. An inverse correlation was observed between Ca, Mg, and Ca/P in hair and the metabolic score. Inverse correlations were also observed for individual metabolic parameters (i.e., diastolic blood pressure, homeostasis model assessment for insulin resistance, non-HDL cholesterol). In particular, girls with a total number of three or more metabolic parameters above the age-specific 75th percentile showed significantly reduced hair Ca, Mg, and Ca/P concentrations. This study showed reduced hair mineral concentrations in young girls with a more unhealthy metabolic profile. Positive associations were observed between some minerals and BMI and BF%.