SEAGLE: A Scalable Exact Algorithm for Large-Scale Set-Based Gene-Environment Interaction Tests in Biobank Data.

The explosion of biobank data offers unprecedented opportunities for gene-environment interaction (GxE) studies of complex diseases because of the large sample sizes and the rich collection in genetic and non-genetic information. However, the extremely large sample size also introduces new computational challenges in G×E assessment, especially for set-based G×E variance component (VC) tests, which are a widely used strategy to boost overall G×E signals and to evaluate the joint G×E effect of multiple variants from a biologically meaningful unit (e.g., gene). In this work, we focus on continuous traits and present SEAGLE, a Scalable Exact AlGorithm for Large-scale set-based G×E tests, to permit G×E VC tests for biobank-scale data. SEAGLE employs modern matrix computations to calculate the test statistic and p-value of the GxE VC test in a computationally efficient fashion, without imposing additional assumptions or relying on approximations. SEAGLE can easily accommodate sample sizes in the order of 105, is implementable on standard laptops, and does not require specialized computing equipment. We demonstrate the performance of SEAGLE using extensive simulations. We illustrate its utility by conducting genome-wide gene-based G×E analysis on the Taiwan Biobank data to explore the interaction of gene and physical activity status on body mass index.

PROBABILISTIC ERROR ANALYSIS FOR INNER PRODUCTS.

Probabilistic models are proposed for bounding the forward error in the numerically computed inner product (dot product, scalar product) between two real n-vectors. We derive probabilistic perturbation bounds as well as probabilistic roundoff error bounds for the sequential accumulation of the inner product. These bounds are nonasymptotic, explicit, with minimal assumptions, and with a clear relationship between failure probability and relative error. The roundoffs are represented as bounded, zero-mean random variables that are independent or have conditionally independent means. Our probabilistic bounds are based on Azuma's inequality and its associated martingale, which mirrors the sequential order of computations. The derivation of forward error bounds "from first principles" has the advantage of producing condition numbers that are customized for the probabilistic bounds. Numerical experiments confirm that our bounds are more informative, often by several orders of magnitude, than traditional deterministic bounds-even for small vector dimensions n and very stringent success probabilities. In particular the probabilistic roundoff error bounds are functions of n rather than n, thus giving a quantitative confirmation of Wilkinson's intuition. The paper concludes with a critical assessment of the probabilistic approach.