Using a collection of simulated an real benchmarks, we compare Bayesian and frequentist regularization approaches under a low informative constraint when the number of variables is almost equal to the number of observations on simulated and real datasets. This comparison includes new global noninformative approaches for Bayesian variable selection built on Zellner's g-priors that are similar to Liang et al. (2008). The interest of those calibration-free proposals is discussed. The numerical experiments we present highlight the appeal of Bayesian regularization methods, when compared with non-Bayesian alternatives. They dominate frequentist methods in the sense that they provide smaller prediction errors while selecting the most relevant variables in a parsimonious way.

Background Current non-invasive methods of assessing disease activity in systemic lupus erythematosus (SLE) are of limited sensitivity and specificity. Testing includes acute phase markers, autoantibodies and complement levels. Although measurements of dsDNA antibodies and complement C3/C4 levels are routine, they remain of limited value. Improved blood and urine markers may help in early detection of flare, distinction between flare and chronic damage, and monitoring response to therapy. Methods A total of 87 patients with SLE were tested for the following cytokines in serum and urine: monocyte chemoattractant protein 1 (MCP-1), regulated upon activation, normal T cell expressed and secreted (RANTES), soluble tumour necrosis factor receptor 1 (sTNF-R1), interferon-inducible protein 10 (IP-10), monocyte inhibitory protein 1α (MIP-1α) and vascular endothelial growth factor (VEGF). Patients attending the Lupus Unit at St Thomas’ Hospital, London, UK were divided into active lupus nephritis (LN), inactive LN and non-renal SLE groups based on their renal pathology and SLE disease activity index (SLEDAI). Cytokine testing was performed using the FIDIS multiplex bead assay. Results The mean level of serum sTNF-R1 was higher in the active LN group compared with both inactive LN and non-renal SLE groups ( p < 0.001). For urine measurements there were significant differences between active LN and non-renal SLE for VEGF ( p = 0.016), after statistical correction for multiple testing. Both urinary and serum sTNF-R1 and IP-10 levels correlated with SLEDAI scores ( p < 0.001), while serum VEGF correlated weakly with SLEDAI ( p = 0.025). The optimum combination for differentiating active from inactive LN patients was serum VEGF, sTNF-R1, MCP-1 and glomerular filtration rate plus urinary sTNF-R1 and protein-creatinine ratio. Conclusion These results indicate that for active LN, sTNF-R1 could be a useful serum cytokine marker, with potential for VEGF in the urine. This study has confirmed the ability of the multiplex bead technique to detect cytokines in a good analytical range, including very low and high levels, in both serum and urine. Combining serum and urine markers provided additional sensitivity in distinguishing active from inactive LN.

Clinical implementation of pharmacogenomics will help in personalizing drug prescriptions and alleviate the personal and financial burden due to inefficacy and adverse reactions to drugs. However, such implementation is lagging in many parts of the world, including the Middle East, mainly due to the lack of data on the distribution of actionable pharmacogenomic variation in these ethnicities. We analyzed 6,045 whole genomes from the Qatari population for the distribution of allele frequencies of 2,629 variants in 1,026 genes known to affect 559 drugs or classes of drugs. We also performed a focused analysis of genotypes or diplotypes of 15 genes affecting 46 drugs, which have guidelines for clinical implementation and predicted their phenotypic impact. The allele frequencies of 1,320 variants in 703 genes affecting 299 drugs or class of drugs were significantly different between the Qatari population and other world populations. On average, Qataris carry 3.6 actionable genotypes/diplotypes, affecting 13 drugs with guidelines for clinical implementation, and 99.5% of the individuals had at least one clinically actionable genotype/diplotype. Increased risk of simvastatin-induced myopathy could be predicted in ~32% of Qataris from the diplotypes of SLCO1B1, which is higher compared to many other populations, while fewer Qataris may need tacrolimus dosage adjustments for achieving immunosuppression based on the CYP3A5 diplotypes compared to other world populations. Distinct distribution of actionable pharmacogenomic variation was also observed among the Qatari subpopulations. Our comprehensive study of the distribution of actionable genetic variation affecting drugs in a Middle Eastern population has potential implications for preemptive pharmacogenomic implementation in the region and beyond.

No abstract

Variable selection in linear regression can be challenging, particularly in situations where a large number of predictors is available with possibly high correlations, such as gene expression data. In this paper we propose a new method called the elastic corr-net to simultaneously select variables and encourage a grouping effect where strongly correlated predictors tend to be in or out of the model together. The method is based on penalized least squares with a penalty function that, like the Lasso penalty, shrinks some coefficients to exactly zero. Additionally, this penalty contains a term which explicitly links strength of penalization to the correlation between predictors. A detailed simulation study in small and high dimensional settings is performed, which illustrates the advantages of our approach in relation to several other possible methods. Finally, we apply the methodology to three real data sets. The key contribution of the elastic corr-net is the identification of setting where the elastic net fails to product good results: in terms of prediction accuracy and estimation error, our empirical study suggests that the elastic corr-net is more adapted than the elastic-net to situations where p ≤ n (the number of variables is less or equal to the sample size). if p n, our method remains competitive and also allows the selection of more than n variables in a new way. The first author is partly supported by the project Maroc-STIC. * INRIA Futurs, Projet select, Université Paris-Sud 11 † Libma laboratory, cadi ayyad university, Morroco La régression pénalisée combinant la norme L 1 et une pénalité tenant compte des corrélations entre les variables Résumé : La sélection de variables peutêtre difficile, en particulier dans les situations où un grand nombre de variables explicatives est disponible, avec la présence possible de corrélationsélevées comme dans le cas des données d'expression génétique. Dans cet article, nous proposons une nouvelle méthode de régression linéaire pénalisée, appelée l'elastic corr-net, pour simultanément estimer les paramètres inconnus et sélectionner les variables importantes. De plus, elle encourage un effet de groupe: les variables fortement corrélées ont tendanceàêtre toutes incluses ou toutes exclues du modèle. La méthode est fondée sur les moindres carrés pénalisés avec une pénalité qui, comme la pénalité L 1 , rétrécit certains coefficients exactement vers zéro. En outre, cette pénalité contient un terme qui lie explicitement la force de pénalisationà la corrélation entre les variables explicatives. Pour montrer les avantages de notre approche par rapport aux méthodes les plus concurrentes, uneétude détaillée de simulation est réalisée en moyenne et grande dimension. Enfin, nous appliquons la méthodologieà trois ensembles de données réelles.Le résultat principal de notre méthode est l'identification du cadre où l'elastic-net est moins performante : en effet, en termes des erreurs de prédiction et d'estimation, notre méthode paraît plus adaptée aux situations du type p ≤ n (le n...