Scaled sparse high-dimensional tests for localizing sequence variants

Deep sequencing studies have been generating high-throughput data for high resolution and comprehensive detection of rare and common genetic variants. Existing association tests are powerful to identify functional variants in large samples. These tests, however, have low power to identify susceptible variants in high-dimensional SNP set, where n (the number of observations) is smaller than or close to m (the size of SNP set under testing). We propose a scaled sparse regression approach for localizing susceptible variants set in high-dimensional SNP sets which are ubiquitous in analyses of deep sequencing data. This approach applies sparse regression with scaled L^p (0<p<1) norm regularization. Under a wide range of simulated scenarios, the proposed approach appropriately controlled Type I error rate. In terms of the detection power of susceptible variants, the proposed approach appeared more powerful than several existing prominent methods. The practical utility of the proposed approach is illustrated by application to real DNA sequence data of Mexican Americans from GAW18.