PIntMF: Penalized Integrative Matrix Factorization method for multi-omics data.

MOTIVATION: It is more and more common to perform multi-omics analyses to explore the genome at diverse levels and not only at a single level. Through integrative statistical methods, multi-omics data have the power to reveal new biological processes, potential biomarkers and subgroups in a cohort. Matrix factorization (MF) is an unsupervised statistical method that allows a clustering of individuals, but also reveals relevant omics variables from the various blocks. RESULTS: Here, we present PIntMF (Penalized Integrative Matrix Factorization), an MF model with sparsity, positivity and equality constraints. To induce sparsity in the model, we used a classical Lasso penalization on variable and individual matrices. For the matrix of samples, sparsity helps in the clustering, while normalization (matching an equality constraint) of inferred coefficients is added to improve interpretation. Moreover, we added an automatic tuning of the sparsity parameters using the famous glmnet package. We also proposed three criteria to help the user to choose the number of latent variables. PIntMF was compared with other state-of-the-art integrative methods including feature selection techniques in both synthetic and real data. PIntMF succeeds in finding relevant clusters as well as variables in two types of simulated data (correlated and uncorrelated). Next, PIntMF was applied to two real datasets (Diet and cancer), and it revealed interpretable clusters linked to available clinical data. Our method outperforms the existing ones on two criteria (clustering and variable selection). We show that PIntMF is an easy, fast and powerful tool to extract patterns and cluster samples from multi-omics data. AVAILABILITY AND IMPLEMENTATION: An R package is available at https://github.com/mpierrejean/pintmf. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Clustering and variable selection evaluation of 13 unsupervised methods for multi-omics data integration.

Recent advances in NGS sequencing, microarrays and mass spectrometry for omics data production have enabled the generation and collection of different modalities of high-dimensional molecular data. The integration of multiple omics datasets is a statistical challenge, due to the limited number of individuals, the high number of variables and the heterogeneity of the datasets to integrate. Recently, a lot of tools have been developed to solve the problem of integrating omics data including canonical correlation analysis, matrix factorization and SM. These commonly used techniques aim to analyze simultaneously two or more types of omics. In this article, we compare a panel of 13 unsupervised methods based on these different approaches to integrate various types of multi-omics datasets: iClusterPlus, regularized generalized canonical correlation analysis, sparse generalized canonical correlation analysis, multiple co-inertia analysis (MCIA), integrative-NMF (intNMF), SNF, MoCluster, mixKernel, CIMLR, LRAcluster, ConsensusClustering, PINSPlus and multi-omics factor analysis (MOFA). We evaluate the ability of the methods to recover the subgroups and the variables that drive the clustering on eight benchmarks of simulation. MOFA does not provide any results on these benchmarks. For clustering, SNF, MoCluster, CIMLR, LRAcluster, ConsensusClustering and intNMF provide the best results. For variable selection, MoCluster outperforms the others. However, the performance of the methods seems to depend on the heterogeneity of the datasets (especially for MCIA, intNMF and iClusterPlus). Finally, we apply the methods on three real studies with heterogeneous data and various phenotypes. We conclude that MoCluster is the best method to analyze these omics data. Availability: An R package named CrIMMix is available on GitHub at https://github.com/CNRGH/crimmix to reproduce all the results of this article.

Where are the missing gene defects in inherited retinal disorders? Intronic and synonymous variants contribute at least to 4% of CACNA1F-mediated inherited retinal disorders.

Inherited retinal disorders (IRD) represent clinically and genetically heterogeneous diseases. To date, pathogenic variants have been identified in ~260 genes. Albeit that many genes are implicated in IRD, for 30-50% of the cases, the gene defect is unknown. These cases may be explained by novel gene defects, by overlooked structural variants, by variants in intronic, promoter or more distant regulatory regions, and represent synonymous variants of known genes contributing to the dysfunction of the respective proteins. Patients with one subgroup of IRD, namely incomplete congenital stationary night blindness (icCSNB), show a very specific phenotype. The major cause of this condition is the presence of a hemizygous pathogenic variant in CACNA1F. A comprehensive study applying direct Sanger sequencing of the gene-coding regions, exome and genome sequencing applied to a large cohort of patients with a clinical diagnosis of icCSNB revealed indeed that seven of the 189 CACNA1F-related cases have intronic and synonymous disease-causing variants leading to missplicing as validated by minigene approaches. These findings highlight that gene-locus sequencing may be a very efficient method in detecting disease-causing variants in clinically well-characterized patients with a diagnosis of IRD, like icCSNB.

Shared Heritability of Blood Pressure and Pulse Wave Velocity: Insights From the STANISLAS Cohort.

BACKGROUND: Pulse wave velocity (PWV) is a marker of arterial stiffness, which is intrinsically highly correlated with blood pressure (BP). However, the interplay of PWV and BP heritability has not been extensively studied. This study aimed to estimate the heritability of PWV and BP and determine the genetic correlation between PWV and BP. METHODS: The heritability of PWV and BP was estimated in 1080 subjects from the STANISLAS (Suivi Temporaire Annuel Non-Invasif de la Santé des Lorrains Assurés Sociaux) cohort with at least one relative using a linear mixed model within one frequentist and one Bayesian framework implemented, respectively, in the Gaston and MCMCglmm R packages. Then their genetic correlations were also estimated. RESULTS: The heritability estimations for PWV were within the same range of the heritability of systolic BP and diastolic BP (23%, 19%, and 27%, respectively). Daytime heritability of BP was higher than nighttime BP. In addition, phenotypic correlations between PWV and systolic BP/diastolic BP were, respectively, 0.34 and 0.23, whereas nonsignificant genetic correlations were 0.08 and 0.22 respectively, indicating that PWV and diastolic BP shared more polygenic codeterminants than PWV and systolic BP. CONCLUSIONS: Our results suggest that the heritability of PWV is >20% and within the same range as BP heritability. It also suggests that the link between PWV and BP goes beyond phenotypic association: PWV and BP (in particular diastolic BP) share common genetic determinants. This genetic interdependence of PWV and BP appears largely polygenic.

Layer myocardial strain is the most heritable echocardiographic trait.

AIMS: Myocardial deformation assessed by strain analysis represents a significant advancement in our assessment of cardiac mechanics. However, whether this variable is genetically heritable or whether all/most of its variability is related to environmental factors is currently unknown. We sought to determine the heritability of echocardiographically determined cardiac mechanics indices in a population setting. METHODS AND RESULTS: A total of 1357 initially healthy subjects (women 51.6%; 48.2 ± 14.1 years) were included in this study from 20-year follow-up after the fourth visit of the longitudinal familial STANISLAS cohort (Lorraine, France). Data were acquired using state-of-the-art cardiac ultrasound equipment, using acquisition and measurement protocols recommended by the EACVI (European Association of Cardiovascular Imaging)/ASE (American Society of Echocardiography)/Industry Task Force. Layer-specific global longitudinal strain (GLS) and global circumferential strain (full-wall, subendocardial, and subepicardial) and conventional structural and functional cardiac parameters and their potential heritability were assessed using restricted maximum likelihood analysis, with genetic relatedness matrix calculated from genome-wide association data. Indices of longitudinal/circumferential myocardial function and left ventricular (LV) ejection fraction had low heritability (ranging from 10% to 20%). Diastolic and standard LV function parameters had moderate heritability (ranging from 20% to 30%) except for end-systolic and end-diastolic volumes (30% and 45%, respectively). In contrast, global longitudinal subendocardial strain (GLSEndo)/global longitudinal subepicardial strain (GLSEpi) ratio had a high level of heritability (65%). Except for GLSEndo/GLSEpi ratio, a large percentage of variance remained unexplained (>50%). CONCLUSIONS: In our population cohort, GLSEndo/GLSEpi ratio had a high level of heritability, whereas other classical and mechanical LV function parameters did not. Given the increasing recognition of GLSEndo/GLSEpi ratio as an early/sensitive imaging biomarker of systolic dysfunction, our results suggest the possible existence of individual genetic predispositions to myocardial decline.

Association and performance of polygenic risk scores for breast cancer among French women presenting or not a familial predisposition to the disease.

BACKGROUND: Three partially overlapping breast cancer polygenic risk scores (PRS) comprising 77, 179 and 313 SNPs have been proposed for European-ancestry women by the Breast Cancer Association Consortium (BCAC) for improving risk prediction in the general population. However, the effect of these SNPs may vary from one country to another and within a country because of other factors. OBJECTIVE: To assess their associated risk and predictive performance in French women from (1) the CECILE population-based case-control study, (2) BRCA1 or BRCA2 (BRCA1/2) pathogenic variant (PV) carriers from the GEMO study, and (3) familial breast cancer cases with no BRCA1/2 PV and unrelated controls from the GENESIS study. RESULTS: All three PRS were associated with breast cancer in all studies, with odds ratios per standard deviation varying from 1.7 to 2.0 in CECILE and GENESIS, and hazard ratios varying from 1.1 to 1.4 in GEMO. The predictive performance of PRS313 in CECILE was similar to that reported in BCAC but lower than that in GENESIS (area under the receiver operating characteristic curve (AUC) = 0.67 and 0.75, respectively). PRS were less performant in BRCA2 and BRCA1 PV carriers (AUC = 0.58 and 0.54 respectively). CONCLUSION: Our results are in line with previous validation studies in the general population and in BRCA1/2 PV carriers. Additionally, we showed that PRS may be of clinical utility for women with a strong family history of breast cancer and no BRCA1/2 PV, and for those carrying a predicted PV in a moderate-risk gene like ATM, CHEK2 or PALB2.

Significant correlation between a set of genetic polymorphisms and a functional brain network revealed by feature selection and sparse Partial Least Squares.

Brain imaging is increasingly recognised as an intermediate phenotype to understand the complex path between genetics and behavioural or clinical phenotypes. In this context, a first goal is to propose methods to identify the part of genetic variability that explains some neuroimaging variability. Classical univariate approaches often ignore the potential joint effects that may exist between genes or the potential covariations between brain regions. In this paper, we propose instead to investigate an exploratory multivariate method in order to identify a set of Single Nucleotide Polymorphisms (SNPs) covarying with a set of neuroimaging phenotypes derived from functional Magnetic Resonance Imaging (fMRI). Recently, Partial Least Squares (PLS) regression or Canonical Correlation Analysis (CCA) have been proposed to analyse DNA and transcriptomics. Here, we propose to transpose this idea to the DNA vs. imaging context. However, in very high-dimensional settings like in imaging genetics studies, such multivariate methods may encounter overfitting issues. Thus we investigate the use of different strategies of regularisation and dimension reduction techniques combined with PLS or CCA to face the very high dimensionality of imaging genetics studies. We propose a comparison study of the different strategies on a simulated dataset first and then on a real dataset composed of 94 subjects, around 600,000 SNPs and 34 functional MRI lateralisation indexes computed from reading and speech comprehension contrast maps. We estimate the generalisability of the multivariate association with a cross-validation scheme and demonstrate the significance of this link, using a permutation procedure. Univariate selection appears to be necessary to reduce the dimensionality. However, the significant association uncovered by this two-step approach combining univariate filtering and L1-regularised PLS suggests that discovering meaningful genetic associations calls for a multivariate approach.

Weak Association Between Genetic Markers of Hyperuricemia and Cardiorenal Outcomes: Insights From the STANISLAS Study Cohort With a 20-Year Follow-Up.

Background Hyperuricemia is associated with poor cardiovascular outcomes, although it is uncertain whether this relationship is causal in nature. This study aimed to: (1) assess the heritability of serum uric acid (SUA) levels, (2) conduct a genome-wide association study on SUA levels, and (3) investigate the association between certain single-nucleotide polymorphisms and target organ damage. Methods and Results The STANISLAS (Suivi Temporaire Annuel Non-Invasif de la Santé des Lorrains Assurés Sociaux) study cohort is a single-center longitudinal cohort recruited between 1993 and 1995 (visit 1), with a last visit (visit 4 [V4]) performed ≈20 years apart. Serum lipid profile, SUA, urinary albumin/creatinine ratio, estimated glomerular filtration rate, 24-hour ambulatory blood pressure monitoring, transthoracic echocardiography, pulse wave velocity, and genotyping for each participant were assessed at V4. A total of 1573 participants were included at V4, among whom 1417 had available SUA data at visit 1. Genome-wide association study results highlighted multiple single-nucleotide polymorphisms on the SLC2A9 gene linked to SUA levels. Carriers of the most associated mutated SLC2A9 allele (rs16890979) had significantly lower SUA levels. Although SUA level at V4 was highly associated with diabetes, prediabetes, higher body mass index, CRP (C-reactive protein) levels, estimated glomerular filtration rate variation (visit 1-V4), carotid intima-media thickness, and pulse wave velocity, rs16890979 was only associated with higher carotid intima-media thickness. Conclusions Our findings demonstrate that rs16890979, a genetic determinant of SUA levels located on the SLC2A9 gene, is associated with carotid intima-media thickness despite significant associations between SUA levels and several clinical outcomes, thereby lending support to the hypothesis of a link between SUA and cardiovascular disease.

Impact of natriuretic peptide polymorphisms on diastolic and metabolic function in a populational cohort: insights from the STANISLAS cohort.

AIMS: Elevated brain natriuretic peptide (BNP) and the N-terminal fragment of its pro-hormone (NT-proBNP) have become established biomarkers for heart failure and are associated with cardiovascular morbidity and mortality. Investigating sources of inter-individual heterogeneity, particularly genetic factors, could help better identify patients at risk of future cardiovascular disease. The aim of this study was to estimate the heritability of circulating NT-proBNP levels, to perform a genome-wide association study (GWAS) and gene-candidate analysis focused on NPPB-NPPA genes on these levels, and to examine their association with cardiovascular or metabolic outcomes. METHODS AND RESULTS: A total of 1555 individuals from the STANISLAS study were included. The heritability of circulating NT-proBNP levels was estimated at 15%, with seven single nucleotide polymorphisms (SNPs) reaching the significant threshold in the GWAS. All above SNPs were located on the same gene cluster constituted of MTHFR, CLCN6, NPPA, NPPB, and C1orf167. NPPA gene expression was also associated with NT-proBNP levels. Moreover, six other SNPs from NPPA-NPPB genes were associated with diastolic function (lateral e' on echocardiography) and metabolic features (glycated haemoglobin). CONCLUSIONS: The heritability of natriuretic peptides appears relatively low (15%) and mainly based on the same gene cluster constituted of MTHFR, CLCN6, NPPA, NPPB, and C1orf167. Natriuretic peptide polymorphisms are associated with natriuretic peptide levels and diastolic function. These results suggest that natriuretic peptide polymorphisms may have an impact in the early stages of cardiovascular and metabolic disease.

Identification of risk loci for primary aldosteronism in genome-wide association studies.

Primary aldosteronism affects up to 10% of hypertensive patients and is responsible for treatment resistance and increased cardiovascular risk. Here we perform a genome-wide association study in a discovery cohort of 562 cases and 950 controls and identify three main loci on chromosomes 1, 13 and X; associations on chromosome 1 and 13 are replicated in a second cohort and confirmed by a meta-analysis involving 1162 cases and 3296 controls. The association on chromosome 13 is specific to men and stronger in bilateral adrenal hyperplasia than aldosterone producing adenoma. Candidate genes located within the two loci, CASZ1 and RXFP2, are expressed in human and mouse adrenals in different cell clusters. Their overexpression in adrenocortical cells suppresses mineralocorticoid output under basal and stimulated conditions, without affecting cortisol biosynthesis. Our study identifies the first risk loci for primary aldosteronism and highlights new mechanisms for the development of aldosterone excess.

Region-specific expression of young small-scale duplications in the human central nervous system.

BACKGROUND: The duplication of genes is one of the main genetic mechanisms that led to the gain in complexity of biological tissue. Although the implication of duplicated gene expression in brain evolution was extensively studied through comparisons between organs, their role in the regional specialization of the adult human central nervous system has not yet been well described. RESULTS: Our work explored intra-organ expression properties of paralogs through multiple territories of the human central nervous system (CNS) using transcriptome data generated by the Genotype-Tissue Expression (GTEx) consortium. Interestingly, we found that paralogs were associated with region-specific expression in CNS, suggesting their involvement in the differentiation of these territories. Beside the influence of gene expression level on region-specificity, we observed the contribution of both duplication age and duplication type to the CNS region-specificity of paralogs. Indeed, we found that small scale duplicated genes (SSDs) and in particular ySSDs (SSDs younger than the 2 rounds of whole genome duplications) were more CNS region-specific than other paralogs. Next, by studying the two paralogs of ySSD pairs, we observed that when they were region-specific, they tend to be specific to the same region more often than for other paralogs, showing the high co-expression of ySSD pairs. The extension of this analysis to families of paralogs showed that the families with co-expressed gene members (i.e. homogeneous families) were enriched in ySSDs. Furthermore, these homogeneous families tended to be region-specific families, where the majority of their gene members were specifically expressed in the same region. CONCLUSIONS: Overall, our study suggests the involvement of ySSDs in the differentiation of human central nervous system territories. Therefore, we show the relevance of exploring region-specific expression of paralogs at the intra-organ level.

Genome-wide haplotype association study in imaging genetics using whole-brain sulcal openings of 16,304 UK Biobank subjects.

Neuroimaging-genetics cohorts gather two types of data: brain imaging and genetic data. They allow the discovery of associations between genetic variants and brain imaging features. They are invaluable resources to study the influence of genetics and environment in the brain features variance observed in normal and pathological populations. This study presents a genome-wide haplotype analysis for 123 brain sulcus opening value (a measure of sulcal width) across the whole brain that include 16,304 subjects from UK Biobank. Using genetic maps, we defined 119,548 blocks of low recombination rate distributed along the 22 autosomal chromosomes and analyzed 1,051,316 haplotypes. To test associations between haplotypes and complex traits, we designed three statistical approaches. Two of them use a model that includes all the haplotypes for a single block, while the last approach considers each haplotype independently. All the statistics produced were assessed as rigorously as possible. Thanks to the rich imaging dataset at hand, we used resampling techniques to assess False Positive Rate for each statistical approach in a genome-wide and brain-wide context. The results on real data show that genome-wide haplotype analyses are more sensitive than single-SNP approach and account for local complex Linkage Disequilibrium (LD) structure, which makes genome-wide haplotype analysis an interesting and statistically sound alternative to the single-SNP counterpart.

Exploring the Link Between Additive Heritability and Prediction Accuracy From a Ridge Regression Perspective.

Genome-Wide Association Studies (GWAS) explain only a small fraction of heritability for most complex human phenotypes. Genomic heritability estimates the variance explained by the SNPs on the whole genome using mixed models and accounts for the many small contributions of SNPs in the explanation of a phenotype. This paper approaches heritability from a machine learning perspective, and examines the close link between mixed models and ridge regression. Our contribution is two-fold. First, we propose estimating genomic heritability using a predictive approach via ridge regression and Generalized Cross Validation (GCV). We show that this is consistent with classical mixed model based estimation. Second, we derive simple formulae that express prediction accuracy as a function of the ratio n p , where n is the population size and p the total number of SNPs. These formulae clearly show that a high heritability does not imply an accurate prediction when p > n. Both the estimation of heritability via GCV and the prediction accuracy formulae are validated using simulated data and real data from UK Biobank.

Circulating plasma proteins and new-onset diabetes in a population-based study: proteomic and genomic insights from the STANISLAS cohort.

OBJECTIVE: Determining the factors associated with new-onset pre-diabetes and type 2 diabetes mellitus (T2D) is important for improving the current prevention strategies and for a better understanding of the disease. DESIGN: To study the factors (clinical, circulating protein and genetic) associated with new onset pre-diabetes and T2D in an initially healthy (without diabetes) populational familial cohort with a long follow-up (STANISLAS cohort). METHODS: A total of 1506 participants attended both the visit 1 and visit 4, separated by ≈20 years. Over 400 proteins, GWAS and genetic associations were studied using models adjusted for potential confounders. Both prospective (V1 to V4) and cross-sectional (V4) analyses were performed. RESULTS: People who developed pre-diabetes (n = 555) and/or T2D (n = 73) were older, had higher BMI, blood pressure, glucose, LDL cholesterol, and lower eGFR. After multivariable selection, PAPP-A (pappalysin-1) was the only circulating protein associated with the onset of both pre-diabetes and T2D with associations persisting at visit 4 (i.e. ≈20 years later). FGF-21 (fibroblast growth factor 21) was a strong prognosticator for incident T2D in the longitudinal analysis, but not in the cross-sectional analysis. The heritability of the circulating PAPP-A was estimated at 44%. In GWAS analysis, the SNP rs634737 was associated with PAPP-A both at V1 and V4. External replication also showed lower levels of PAPP-A in patients with T2D. CONCLUSIONS: The risk of developing pre-diabetes and T2D increases with age and with features of the metabolic syndrome. Circulating PAPP-A, which has an important genetic component, was associated with both the development and presence of pre-diabetes and T2D.

Toxoplasma gondii exposure may modulate the influence of TLR2 genetic variation on bipolar disorder: a gene-environment interaction study.

BACKGROUND: Genetic vulnerability to environmental stressors is yet to be clarified in bipolar disorder (BD), a complex multisystem disorder in which immune dysfunction and infectious insults seem to play a major role in the pathophysiology. Association between pattern-recognition receptor coding genes and BD had been previously reported. However, potential interactions with history of pathogen exposure are yet to be explored. METHODS: 138 BD patients and 167 healthy controls were tested for serostatus of Toxoplasma gondii, CMV, HSV-1 and HSV-2 and genotyped for TLR2 (rs4696480 and rs3804099), TLR4 (rs1927914 and rs11536891) and NOD2 (rs2066842) polymorphisms (SNPs). Both the pathogen-specific seroprevalence and the TLR/NOD2 genetic profiles were compared between patients and controls followed by modelling of interactions between these genes and environmental infectious factors in a regression analysis. RESULTS: First, here again we observed an association between BD and Toxoplasma gondii (p = 0.045; OR = 1.77; 95 % CI 1.01-3.10) extending the previously published data on a cohort of a relatively small number of patients (also included in the present sample). Second, we found a trend for an interaction between the TLR2 rs3804099 SNP and Toxoplasma gondii seropositivity in conferring BD risk (p = 0.017, uncorrected). CONCLUSIONS: Pathogen exposure may modulate the influence of the immunogenetic background on BD. A much larger sample size and information on period of pathogen exposure are needed in future gene-environment interaction studies.

Voxelwise multivariate statistics and brain-wide machine learning using the full diffusion tensor.

In this paper, we propose to use the full diffusion tensor to perform brain-wide score prediction on diffusion tensor imaging (DTI) using the log-Euclidean framework., rather than the commonly used fractional anisotropy (FA). Indeed, scalar values such as the FA do not capture all the information contained in the diffusion tensor. Additionally, full tensor information is included in every step of the pre-processing pipeline: registration, smoothing and feature selection using voxelwise multivariate regression analysis. This approach was tested on data obtained from 30 children and adolescents with autism spectrum disorder and showed some improvement over the FA-only analysis.