Biology subjects

Greenwood, C. M.

Publications and source records attributed to Greenwood, C. M..

4 recordsLinked to original sources

bioRxiv2018

Sparse additive interaction learning

A conceptual paradigm for onset of a new disease is often considered to be the result of changes in entire biological networks whose states are affected by a complex interaction of genetic and environmental factors. However, when modelling a relevant phenotype as a function of high dimensional measurements, power to estimate interactions is low, the number of possible interactions could be enormous and their effects may be non-linear. In this work, we introduce a method called sail for detecting non-linear interactions with a key environmental or exposure variable in high-dimensional settings which respects the strong or weak heredity constraints. We prove that asymptotically, our method possesses the oracle property, i.e., it performs as well as if the true model were known in advance. We develop a computationally efficient fitting algorithm with automatic tuning parameter selection, which scales to high-dimensional datasets. Through an extensive simulation study, we show that sail outperforms existing penalized regression methods in terms of prediction accuracy and support recovery when there are non-linear interactions with an exposure variable. We apply sail to detect non-linear interactions between genes and a prenatal psychosocial intervention program on cognitive performance in children at 4 years of age. Results show that individuals who are genetically predisposed to lower educational attainment are those who stand to benefit the most from the intervention. Our algorithms are implemented in an R package available on CRAN (https://cran.r-project.org/package=sail).

bioinformatics↗

bioRxiv2018

Machine Learning to Predict Osteoporotic Fracture Risk from Genotypes

BackgroundGenomics-based prediction could be useful since genome-wide genotyping costs less than many clinical tests. We tested whether machine learning methods could provide a clinically-relevant genomic prediction of quantitative ultrasound speed of sound (SOS)--a risk factor for osteoporotic fracture.\n\nMethodsWe used 341,449 individuals from UK Biobank with SOS measures to develop genomically-predicted SOS (gSOS) using machine learning algorithms. We selected the optimal algorithm in 5,335 independent individuals and then validated it and its ability to predict incident fracture in an independent test dataset (N = 80,027). Finally, we explored whether genomic pre-screening could complement a UK-based osteoporosis screening strategy, based on the validated tool FRAX.\n\nResultsgSOS explained 4.8-fold more variance in SOS than FRAX clinical risk factors (CRF) alone (r2 = 23% vs. 4.8%). A standard deviation decrease in gSOS, adjusting for the CRF-FRAX score was associated with a higher increased odds of incident major osteoporotic fracture (1,491 cases / 78,536 controls, OR = 1.91 [1.70-2.14], P = 10-28) than that for measured SOS (OR = 1.60 [1.50-1.69], P = 10-52) and femoral neck bone mineral density (147 cases / 4,594 controls, OR = 1.53 [1.27-1.83], P = 10-6). Individuals in the bottom decile of the gSOS distribution had a 3.25-fold increased risk of major osteoporotic fracture (P = 10-18) compared to the top decile. A gSOS-based FRAX score, identified individuals at high risk for incident major osteoporotic fractures better than the CRF-FRAX score (P = 10-14). Introducing a genomic pre-screening step into osteoporosis screening in 4,741 individuals reduced the number of required clinical visits from 2,455 to 1,273 and the number of BMD tests from 1,013 to 473, while only reducing the sensitivity to identify individuals eligible for therapy from 99% to 95%.\n\nInterpretationThe use of genotypes in a machine learning algorithm resulted in a clinically-relevant prediction of SOS and fracture, with potential to impact healthcare resource utilization.\n\nResearch in ContextO_ST_ABSEvidence Before this StudyC_ST_ABSGenome-wide association studies have identified many loci associated with risk of clinically-relevant fracture risk factors, such as SOS. Yet, it is unclear if such information can be leveraged to identify those at risk for disease outcomes, such as osteoporotic fractures. Most previous attempts to predict disease risk from genotypes have used polygenic risk scores, which may not be optimal for genomic-prediction. Despite these obstacles, genomic-prediction could enable screening programs to be more efficient since most people screened in a population are not determined to have a level of risk that would prompt a change in clinical care. Genomic pre-screening could help identify individuals whose risk of disease is low enough that they are unlikely to benefit from screening.\n\nAdded Value of this StudyUsing a large dataset of 426,811 individuals we trained and tested a machine learning algorithm to genomically-predict SOS. This metric, gSOS, had performance characteristics for predicting fracture risk that were similar to measured SOS and femoral neck BMD. Implementing a gSOS-based pre-screening step into the UK-based osteoporosis treatment guidelines reduced the number of individuals who would require screening clinical visits and skeletal testing by approximately 50%, while having little impact on the sensitivity to identify individuals at high risk for osteoporotic fracture.\n\nImplications of all of the Available EvidenceClinically-relevant genomic prediction of heritable traits is feasible using the machine learning algorithm presented here in large sample sizes. Genome-wide genotyping is now less expensive than many clinical tests, needs to be performed once over a lifetime and could risk stratify for multiple heritable traits and diseases years prior to disease onset, providing an opportunity for prevention. The implementation of such algorithms could improve screening efficiency, yet their cost-effectiveness will need to be ascertained in subsequent analyses.

genomics↗

bioRxiv2018

A General Framework for Variable Selection in Linear Mixed Models with Applications to Genetic Studies with Structured Populations

Complex traits are known to be influenced by a combination of environmental factors and rare and common genetic variants. However, detection of such multivariate associations can be compromised by low statistical power and confounding by population structure. Linear mixed effects models (LMM) can account for correlations due to relatedness but have not been applicable in high-dimensional (HD) settings where the number of fixed effect predictors greatly exceeds the number of samples. False positives or false negatives can result from two-stage approaches, where the residuals estimated from a null model adjusted for the subjects relationship structure are sub-sequently used as the response in a standard penalized regression model. To overcome these challenges, we develop a general penalized LMM with a single random effect called ggmix for simultaneous SNP selection and adjustment for population structure in high dimensional prediction models. We develop a blockwise coordinate descent algorithm with automatic tuning parameter selection which is highly scalable, computationally efficient and has theoretical guarantees of convergence. Through simulations and three real data examples, we show that ggmix leads to more parsimonious models compared to the two-stage approach or principal component adjustment with better prediction accuracy. Our method performs well even in the presence of highly correlated markers, and when the causal SNPs are included in the kinship matrix. ggmix can be used to construct polygenic risk scores and select instrumental variables in Mendelian randomization studies. Our algorithms are available in an R package (https://github.com/greenwoodlab/ggmix). 1 Author SummaryThis work addresses a recurring challenge in the analysis and interpretation of genetic association studies: which genetic variants can best predict and are independently associated with a given phenotype in the presence of population structure ? Not controlling confounding due to geographic population structure, family and/or cryptic relatedness can lead to spurious associations. Much of the existing research has therefore focused on modeling the association between a phenotype and a single genetic variant in a linear mixed model with a random effect. However, this univariate approach may miss true associations due to the stringent significance thresholds required to reduce the number of false positives and also ignores the correlations between markers. We propose an alternative method for fitting high-dimensional multivariable models, which selects SNPs that are independently associated with the phenotype while also accounting for population structure. We provide an efficient implementation of our algorithm and show through simulation studies and real data examples that our method outperforms existing methods in terms of prediction accuracy and controlling the false discovery rate.

bioinformatics↗

bioRxiv2017

An analytic approach for interpretable predictive models in high dimensional data, in the presence of interactions with exposures

Predicting a phenotype and understanding which variables improve that prediction are two very challenging and overlapping problems in analysis of high-dimensional data such as those arising from genomic and brain imaging studies. It is often believed that the number of truly important predictors is small relative to the total number of variables, making computational approaches to variable selection and dimension reduction extremely important. To reduce dimensionality, commonly-used two-step methods first cluster the data in some way, and build models using cluster summaries to predict the phenotype.\n\nIt is known that important exposure variables can alter correlation patterns between clusters of high-dimensional variables, i.e., alter network properties of the variables. However, it is not well understood whether such altered clustering is informative in prediction. Here, assuming there is a binary exposure with such network-altering effects, we explore whether use of exposure-dependent clustering relationships in dimension reduction can improve predictive modelling in a two-step framework. Hence, we propose a modelling framework called ECLUST to test this hypothesis, and evaluate its performance through extensive simulations.\n\nWith ECLUST, we found improved prediction and variable selection performance compared to methods that do not consider the environment in the clustering step, or to methods that use the original data as features. We further illustrate this modelling framework through the analysis of three data sets from very different fields, each with high dimensional data, a binary exposure, and a phenotype of interest. Our method is available in the eclust CRAN package.

genomics↗