The widespread availability of high-dimensional biological data has made the simultaneous screening of numerous biological characteristics a central statistical problem in computational biology. While the dimensionality of such datasets continues to increase, the problem of teasing out the effects of biomarkers in studies measuring baseline confounders while avoiding model misspecification remains only partially addressed. Efficient estimators constructed from data adaptive estimates of the data-generating distribution provide an avenue for avoiding model misspecification; however, in the context of high-dimensional problems requiring simultaneous estimation of numerous parameters, standard variance estimators have proven unstable, resulting in unreliable Type-I error control under standard multiple testing corrections. We present the formulation of a general approach for applying empirical Bayes shrinkage approaches to asymptotically linear estimators of parameters defined in the nonparametric model. The proposal applies existing shrinkage estimators to the estimated variance of the influence function, allowing for increased inferential stability in high-dimensional settings. A methodology for nonparametric variable importance analysis for use with high-dimensional biological datasets with modest sample sizes is introduced and the proposed technique is demonstrated to be robust in small samples even when relying on data adaptive estimators that eschew parametric forms. Use of the proposed variance moderation strategy in constructing stabilized variable importance measures of biomarkers is demonstrated by application to an observational study of occupational exposure. The result is a data adaptive approach for robustly uncovering stable associations in high-dimensional data with limited sample sizes.
تحميل البحث