Smoking Gun or Circumstantial Evidence? Comparison of Statistical Learning Methods using Functional Annotations for Prioritizing Risk Variants

Reaching the End-Game for GWAS: Machine Learning Approaches for the Prioritization of Complex Disease Loci

Genome-wide association studies (GWAS) have revealed thousands of genetic loci that underpin the complex biology of many human traits. However, the strength of GWAS - the ability to detect genetic association by linkage disequilibrium (LD) - is also its limitation. Whilst the ever-increasing study size and improved design have augmented the power of GWAS to detect effects, differentiation of causal variants or genes from other highly correlated genes associated by LD remains the real challenge. This has severely hindered the biological insights and clinical translation of GWAS findings. Although thousands of disease susceptibility loci have been reported, causal genes at these loci remain elusive. Machine learning (ML) techniques offer an opportunity to dissect the heterogeneity of variant and gene signals in the post-GWAS analysis phase. ML models for GWAS prioritization vary greatly in their complexity, ranging from relatively simple logistic regression approaches to more complex ensemble models such as random forests and gradient boosting, as well as deep learning models, i.e., neural networks. Paired with functional validation, these methods show important promise for clinical translation, providing a strong evidence-based approach to direct post-GWAS research. However, as ML approaches continue to evolve to meet the challenge of causal gene identification, a critical assessment of the underlying methodologies and their applicability to the GWAS prioritization problem is needed. This review investigates the landscape of ML applications in three parts: selected models, input features, and output model performance, with a focus on prioritizations of complex disease associated loci. Overall, we explore the contributions ML has made towards reaching the GWAS end-game with consequent wide-ranging translational impact.

A benchmark study of scoring methods for non-coding mutations

Motivation

Detailed knowledge of coding sequences has led to different candidate models for pathogenic variant prioritization. Several deleteriousness scores have been proposed for the non-coding part of the genome, but no large-scale comparison has been realized to date to assess their performance.

Results

We compared the leading scoring tools (CADD, FATHMM-MKL, Funseq2 and GWAVA) and some recent competitors (DANN, SNP and SOM scores) for their ability to discriminate assumed pathogenic variants from assumed benign variants (using the ClinVar, COSMIC and 1000 genomes project databases). Using the ClinVar benchmark, CADD was the best tool for detecting the pathogenic variants that are mainly located in protein coding gene regions. Using the COSMIC benchmark, FATHMM-MKL, GWAVA and SOMliver outperformed the other tools for pathogenic variants that are typically located in lincRNAs, pseudogenes and other parts of the non-coding genome. However, all tools had low precision, which could potentially be improved by future non-coding genome feature discoveries. These results may have been influenced by the presence of potential benign variants in the COSMIC database. The development of a gold standard as consistent as ClinVar for these regions will be necessary to confirm our tool ranking.

Availability and implementation

The Snakemake, C++ and R codes are freely available from https://github.com/Oncostat/BenchmarkNCVTools and supported on Linux.

Contact

damien.drubay@gustaveroussy.fr or stefan.michiels@gustaveroussy.fr.

Supplementary information

Supplementary data are available at Bioinformatics online.

Collective interaction effects associated with mammalian behavioral traits reveal genetic factors connecting fear and hemostasis

BACKGROUND

Investigation of the genetic architectures that influence the behavioral traits of animals can provide important insights into human neuropsychiatric phenotypes. These traits, however, are often highly polygenic, with individual loci contributing only small effects to the overall association. The polygenicity makes it challenging to explain, for example, the widely observed comorbidity between stress and cardiac disease.

METHODS

We present an algorithm for inferring the collective association of a large number of interacting gene variants with a quantitative trait. Using simulated data, we demonstrate that by taking into account the non-uniform distribution of genotypes within a cohort, we can achieve greater power than regression-based methods for high-dimensional inference.

RESULTS

We analyzed genome-wide data sets of outbred mice and pet dogs, and found neurobiological pathways whose associations with behavioral traits arose primarily from interaction effects: γ-carboxylated coagulation factors and downstream neuronal signaling were highly associated with conditioned fear, consistent with our previous finding in human post-traumatic stress disorder (PTSD) data. Prepulse inhibition in mice was associated with serotonin transporter and platelet homeostasis, and noise-induced fear in dogs with hemostasis.

CONCLUSIONS

Our findings suggest a novel explanation for the observed comorbidity between PTSD/anxiety and cardiovascular diseases: key coagulation factors modulating hemostasis also regulate synaptic plasticity affecting the learning and extinction of fear.

Leveraging cell type specific regulatory regions to detect SNPs associated with tissue factor pathway inhibitor plasma levels

Tissue factor pathway inhibitor (TFPI) regulates the formation of intravascular blood clots, which manifest clinically as ischemic heart disease, ischemic stroke, and venous thromboembolism (VTE). TFPI plasma levels are heritable, but the genetics underlying TFPI plasma level variability are poorly understood. Herein we report the first genome-wide association scan (GWAS) of TFPI plasma levels, conducted in 251 individuals from five extended French-Canadian Families ascertained on VTE. To improve discovery, we also applied a hypothesis-driven (HD) GWAS approach that prioritized single nucleotide polymorphisms (SNPs) in (1) hemostasis pathway genes, and (2) vascular endothelial cell (EC) regulatory regions, which are among the highest expressers of TFPI. Our GWAS identified 131 SNPs with suggestive evidence of association (P-value < 5 × 10^-8 ), but no SNPs reached the genome-wide threshold for statistical significance. Hemostasis pathway genes were not enriched for TFPI plasma level associated SNPs (global hypothesis test P-value = 0.147), but EC regulatory regions contained more TFPI plasma level associated SNPs than expected by chance (global hypothesis test P-value = 0.046). We therefore stratified our genome-wide SNPs, prioritizing those in EC regulatory regions via stratified false discovery rate (sFDR) control, and reranked the SNPs by q-value. The minimum q-value was 0.27, and the top-ranked SNPs did not show association evidence in the MARTHA replication sample of 1,033 unrelated VTE cases. Although this study did not result in new loci for TFPI, our work lays out a strategy to utilize epigenomic data in prioritization schemes for future GWAS studies.

Rapporteur summaries of plenary, symposia, and oral sessions from the XXIIIrd World Congress of Psychiatric Genetics Meeting in Toronto, Canada, 16-20 October 2015

The XXIIIrd World Congress of Psychiatric Genetics meeting, sponsored by the International Society of Psychiatric Genetics, was held in Toronto, ON, Canada, on 16-20 October 2015. Approximately 700 participants attended to discuss the latest state-of-the-art findings in this rapidly advancing and evolving field. The following report was written by trainee travel awardees. Each was assigned one session as a rapporteur. This manuscript represents the highlights and topics that were covered in the plenary sessions, symposia, and oral sessions during the conference, and contains major notable and new findings.

Assessing models for genetic prediction of complex traits: a comparison of visualization and quantitative methods

Background

In silico models have recently been created in order to predict which genetic variants are more likely to contribute to the risk of a complex trait given their functional characteristics. However, there has been no comprehensive review as to which type of predictive accuracy measures and data visualization techniques are most useful for assessing these models.

Methods

We assessed the performance of the models for predicting risk using various methodologies, some of which include: receiver operating characteristic (ROC) curves, histograms of classification probability, and the novel use of the quantile-quantile plot. These measures have variable interpretability depending on factors such as whether the dataset is balanced in terms of numbers of genetic variants classified as risk variants versus those that are not.

Results

We conclude that the area under the curve (AUC) is a suitable starting place, and for models with similar AUCs, violin plots are particularly useful for examining the distribution of the risk scores.

Electronic supplementary material

The online version of this article (doi:10.1186/s12864-015-1616-z) contains supplementary material, which is available to authorized users.