Regional heritability advanced complex trait analysis for GPU and traditional parallel architectures

MOTIVATION

Quantification of the contribution of genetic variation to phenotypic variation for complex traits becomes increasingly computationally demanding with increasing numbers of single-nucleotide polymorphisms and individuals. To meet the challenges in making feasible large-scale studies, we present the REgional heritability advanced complex trait analysis software. Adapted from advanced complex trait analysis (and, in turn, genome-wide complex trait analysis), it is tailored to exploit the parallelism present in modern traditional and graphics processing unit (GPU)-accelerated machines, from workstations to supercomputers.

RESULTS

We adapt the genetic relationship matrix estimation algorithm to remove limitations on memory, allowing the analysis of large datasets. We build on this to develop a version of the code able to efficiently exploit GPU-accelerated systems for both the genetic relationship matrix and REstricted maximum likelihood (REML) parts of the analysis, offering substantial speedup over the traditional central processing unit version. We develop the ability to analyze multiple small regions of the genome across multiple compute nodes in parallel, following the 'regional heritability' approach. We demonstrate the new software using 1024 GPUs in parallel on one of the world's fastest supercomputers.

AVAILABILITY

The code is freely available at http://www.epcc.ed.ac.uk/software-products CONTACT: a.gray@ed.ac.uk.

Prediction of mortality and macrovascular complications in type 2 diabetes: validation of the UKPDS Outcomes Model in the Casale Monferrato Survey, Italy

AIMS/HYPOTHESIS

The United Kingdom Prospective Diabetes Study (UKPDS) Outcomes Model can be used to estimate the lifetime occurrence of major diabetes-related complications in order to calculate health economic outcomes. The aim of the study was to assess the performance of the model by comparing the predicted and observed mortality and the incidence of macrovascular complications in an Italian population-based cohort with type 2 diabetes.

METHODS

We used data from the Casale Monferrato Survey, a cohort enrolled in 1988 and surveyed in 1991 (n = 1,967) to assess the prevalence of cardiovascular risk factors. In 2000, a new survey included all the members of the original cohort who were still alive (n = 860), and in addition all individuals identified with a new diagnosis of type 2 diabetes since 1993 (n = 2,389). We compared the mortality predicted by the model for the 1991 survey over the subsequent 17-year period with the observed risk. The following outcomes were analysed in the 2000 survey: myocardial infarction (MI), other ischaemic heart disease, stroke, congestive heart failure (CHF) and amputation.

RESULTS

For all-cause mortality, the predictions from the model at 5 and 10 years (23% and 47%, respectively) were identical to the observed risks. At 15 years, the risk of death was slightly overestimated (an estimate of 67% vs 64% observed, 95% CI 61%, 66%). The performance of the model was best for patients with a recent history of disease (duration <6 years). Among the complications, the predicted cumulative incidences of MI and CHF were very close to those observed.

CONCLUSIONS/INTERPRETATION

External validation is essential to assess the accuracy of simulation models. The UKPDS Outcomes Model satisfactorily predicted a set of actual incidences of mortality and complications in an Italian diabetes cohort up to a duration of approximately 12 years. The longer term performance of such models should be carefully evaluated.

Five-year results of a randomised controlled trial comparing mobile and fixed bearings in total knee replacement

There is conflicting evidence about the merits of mobile bearings in total knee replacement, partly because most randomised controlled trials (RCTs) have not been adequately powered. We report the results of a multicentre RCT of mobile versus fixed bearings. This was part of the knee arthroplasty trial (KAT), where 539 patients were randomly allocated to mobile or fixed bearings and analysed on an intention-to-treat basis. The primary outcome measure was the Oxford Knee Score (OKS) plus secondary measures including Short Form-12, EuroQol EQ-5D, costs, cost-effectiveness and need for further surgery. There was no significant difference between the groups pre-operatively: mean OKS was 17.18 (sd 7.60) in the mobile-bearing group and 16.49 (sd 7.40) in the fixed-bearing group. At five years mean OKS was 33.19 (sd 16.68) and 33.65 (sd 9.68), respectively. There was no significant difference between trial groups in OKS at five years (-1.12 (95% confidence interval -2.77 to 0.52) or any of the other outcome measures. Furthermore, there was no significant difference in the proportion of patients with knee-related re-operations or in total costs. In this appropriately powered RCT, over the first five years after total knee replacement functional outcomes, re-operation rates and healthcare costs appear to be the same irrespective of whether a mobile or fixed bearing is used.

Advanced complex trait analysis

MOTIVATION

The Genome-wide Complex Trait Analysis (GCTA) software package can quantify the contribution of genetic variation to phenotypic variation for complex traits. However, as those datasets of interest continue to increase in size, GCTA becomes increasingly computationally prohibitive. We present an adapted version, Advanced Complex Trait Analysis (ACTA), demonstrating dramatically improved performance.

RESULTS

We restructure the genetic relationship matrix (GRM) estimation phase of the code and introduce the highly optimized parallel Basic Linear Algebra Subprograms (BLAS) library combined with manual parallelization and optimization. We introduce the Linear Algebra PACKage (LAPACK) library into the restricted maximum likelihood (REML) analysis stage. For a test case with 8999 individuals and 279,435 single nucleotide polymorphisms (SNPs), we reduce the total runtime, using a compute node with two multi-core Intel Nehalem CPUs, from ∼17 h to ∼11 min.

AVAILABILITY AND IMPLEMENTATION

The source code is fully available under the GNU Public License, along with Linux binaries. For more information see http://www.epcc.ed.ac.uk/software-products/acta.

CONTACT

a.gray@ed.ac.uk

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Enabling HIV self-testing in South Africa

In a South African context, we consider the implications of the United States Food and Drug Administration’s recent approval of the OraQuick HIV self-testing kit. We argue that current law and policy inhibit the roll-out of accurate and well-regulated self-testing kits, and create a loophole for sale in supermarkets, but not pharmacies.

Predictors of fracture following suspected injury to the scaphoid

A prospective study was performed to develop a clinical prediction rule that incorporated demographic and clinical factors predictive of a fracture of the scaphoid. Of 260 consecutive patients with a clinically suspected or radiologically confirmed scaphoid fracture, 223 returned for evaluation two weeks after injury and formed the basis of our analysis. Patients were evaluated within 72 hours of injury and at approximately two and six weeks after injury using clinical assessment and standard radiographs. Demographic data and the results of seven specific tests in the clinical examination were recorded. There were 116 (52%) men and their mean age was 33 years (13 to 95; SD 17.9). In 62 patients (28%) a scaphoid fracture was confirmed. A logistic regression model identified male gender (p = 0.002), sports injury (p = 0.004), anatomical snuff box pain on ulnar deviation of the wrist within 72 hours of injury (p < 0.001), and scaphoid tubercle tenderness at two weeks (p < 0.001) as independent predictors of fracture. All patients with no pain at the anatomical snuff box on ulnar deviation of the wrist within 72 hours of injury did not have a fracture (n = 72, 32%). With four independently significant factors positive, the risk of fracture was 91%. Our study has demonstrated that clinical prediction rules have a considerable influence on the probability of a suspected scaphoid fracture. This will help improve the use of supplementary investigations where the diagnosis remains in doubt.

Controlling the topology of Fermi surfaces in metal nanofilms

The properties of metal crystals are governed by the electrons of the highest occupied states at the Fermi level and determined by Fermi surfaces, the Fermi energy contours in momentum space. Topological regulation of the Fermi surface has been an important issue in synthesizing functional materials, which we found to be realized at room temperature in nanometer-thick films. Reducing the thickness of a metal thin film down to its electron wavelength scale induces the quantum size effect and the electronic system changes from three to two-dimensional, transforming the Fermi surface topology. Such an ultrathin film further changes its topology through one-dimensional (1D) structural deformation of the film when it is grown on a 1D substrate. In particular, when the interface has 1D metallic bands, the system is additionally stabilized by forming an electron energy gap by hybridization between 1D states of the film and substrate.

A national analysis of complications following hemiarthroplasty for hip fracture in older patients

BACKGROUND

There is emerging evidence that patients with fractured neck of femur (FNOF) aged >85 years have different demands on a health-care system when compared to younger patients.

AIM

We sought to better quantify this in terms of comorbidity and complication rates.

DESIGN

Retrospective review of national database.

METHODS

Data on all patients who underwent hip hemiarthroplasty for FNOF between January 2005 and December 2008 were extracted from the English hospital episode statistics database.

RESULTS

There were 41 770 patients aged 65-84 years and 35 321 patients aged ≥85 years. The older cohort was less likely to have diabetes, chronic obstructive pulmonary disease and rheumatoid arthritis. However, they exhibited a significantly higher risk of lower respiratory tract infection [odds ratio (OR) = 1.58, 95% confidence interval (CI) 1.50-1.67)], myocardial infarction (OR = 1.67, 1.52-1.83) and acute renal failure (OR = 1.54, 1.40-1.70) within 30 days of surgery with an inpatient mortality risk at 90 days, double that of the younger age group. Length of stay (LoS) was significantly longer in patients >85 years compared to younger patients (median 18 days vs. 15, P < 0.001).

CONCLUSION

Patients aged ≥85 years admitted for FNOF were found to have a lower incidence of major chronic disease but exhibited a greater incidence of acute events following hemiarthroplasty and their LoS was increased. Targeted medical interventions that focus upon this susceptible patient group may help reduce morbidity and improve survival.