Robust inference with knockoffs

DOI10.1214/19-AOS1852MaRDI QIDQ2196226zbMATH OpenOpenAlexFDO

Authors Rina Foygel Barber, Emmanuel J. Candès, Richard Samworth

Publication date 28 August 2020

Published in The Annals of Statistics (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1801.03896, https://projecteuclid.org/euclid.aos/1594972823

zbMATH Keywords

false discovery ratehigh-dimensional regressionknockoffsrobustnessvariable selection

Mathematics Subject Classification ID

Parametric hypothesis testing (62F03) Robustness and adaptive procedures (parametric inference) (62F35) Nonparametric hypothesis testing (62G10) Nonparametric robustness (62G35)

Abstract: We consider the variable selection problem, which seeks to identify important variables influencing a response $Y$ out of many candidate features $X_1,ldots,X_p$. We wish to do so while offering finite-sample guarantees about the fraction of false positives - selected variables $X_j$ that in fact have no effect on $Y$ after the other features are known. When the number of features $p$ is large (perhaps even larger than the sample size $n$), and we have no prior knowledge regarding the type of dependence between $Y$ and $X$, the model-X knockoffs framework nonetheless allows us to select a model with a guaranteed bound on the false discovery rate, as long as the distribution of the feature vector $X=(X_1,dots,X_p)$ is exactly known. This model selection procedure operates by constructing "knockoff copies'" of each of the $p$ features, which are then used as a control group to ensure that the model selection algorithm is not choosing too many irrelevant features. In this work, we study the practical setting where the distribution of $X$ could only be estimated, rather than known exactly, and the knockoff copies of the $X_j$'s are therefore constructed somewhat incorrectly. Our results, which are free of any modeling assumption whatsoever, show that the resulting model selection procedure incurs an inflation of the false discovery rate that is proportional to our errors in estimating the distribution of each feature $X_j$ conditional on the remaining features $X_k:keqj$. The model-X knockoff framework is therefore robust to errors in the underlying assumptions on the distribution of $X$, making it an effective method for many practical applications, such as genome-wide association studies, where the underlying distribution on the features $X_1,dots,X_p$ is estimated accurately but not known exactly.

Review text

There is a research in the domain of variable selection. If the variables \(X_1,\dots,X_p\) provide information about a response $Y$, the goal is to identify the important ones, i.e., the smallest set $S\subseteq\{1,\dots,p\}$ such that $Y$ is independent of the information from $\{(X_j)\}_{j\notin S}$. The set $S$ is known under the name of Markov blanket of $Y$. The focus in the article is on those variable selection methods that could control the false discovery rate, FDR. The authors refer to the ``model-X knockoffs approach developed 2018, in the paper by \textit{E. Candès} et al. [J. R. Stat. Soc., Ser. B, Stat. Methodol. 80, No. 3, 551--577 (2018; Zbl 1398.62335)]. The purpose is to investigate FDR control when running the knockoff filter with estimates of the distribution on $X$ instead of the true distribution \(P_X\). The second section is devoted to the introduction of basics on robust inference with knockoffs, the third section to the presentation of FDR control results and the fourth section to some concrete examples. The proofs of all results are contained in Appendix A.

Recommendations

• Relaxing the assumptions of knockoffs by conditioning
• Panning for Gold: ‘Model-X’ Knockoffs for High Dimensional Controlled Variable Selection
• Controlling the false discovery rate via knockoffs
• Derandomizing Knockoffs
• Powerful knockoffs via minimizing reconstructability
• A pseudo knockoff filter for correlated features
• Adjusting the Benjamini–Hochberg method for controlling the false discovery rate in knockoff-assisted variable selection
• Metropolized Knockoff Sampling
• Gene hunting with hidden Markov model knockoffs
• A high-dimensional power analysis of the conditional randomization test and knockoffs

Cites work

• A dynamic programming algorithm for haplotype block partitioning
• A knockoff filter for high-dimensional selective inference
• Controlling the false discovery rate via knockoffs
• Gene hunting with hidden Markov model knockoffs
• High-dimensional covariance estimation by minimizing \(\ell _{1}\)-penalized log-determinant divergence
• scientific article; zbMATH DE number 720689 (Why is no real title available?)
• Model selection and estimation in the Gaussian graphical model
• On the Benjamini-Hochberg method
• Panning for Gold: ‘Model-X’ Knockoffs for High Dimensional Controlled Variable Selection
• RANK: Large-Scale Inference With Graphical Nonlinear Knockoffs
• Sparse inverse covariance estimation with the graphical lasso
• Strong Control, Conservative Point Estimation and Simultaneous Conservative Consistency of False Discovery Rates: A Unified Approach
• The control of the false discovery rate in multiple testing under dependency.

Cited in

(57)

• A recipe for robust estimation using pseudo data
• Kernel Knockoffs Selection for Nonparametric Additive Models
• False Discovery Rate Control Under General Dependence By Symmetrized Data Aggregation
• Statistical inference for structured high-dimensional models. Abstracts from the workshop held March 11--17, 2018
• Reproducible feature selection in high-dimensional accelerated failure time models
• Reproducible learning in large-scale graphical models
• New perspectives on knockoffs construction
• On the power of conditional independence testing under model-X
• Powerful knockoffs via minimizing reconstructability
• Null-free false discovery rate control using decoy permutations
• Feature screening and FDR control with knockoff features for ultrahigh-dimensional right-censored data
• Relaxing the assumptions of knockoffs by conditioning
• A knockoff filter for high-dimensional selective inference
• Robust AIC with high breakdown scale estimate
• Multilayer knockoff filter: controlled variable selection at multiple resolutions
• Familywise error rate control via knockoffs
• RANK: Large-Scale Inference With Graphical Nonlinear Knockoffs
• A pseudo knockoff filter for correlated features
• A prototype knockoff filter for group selection with FDR control
• Identifying important predictors in large data bases -- multiple testing and model selection
• IPAD: stable interpretable forecasting with knockoffs inference
• Deep knockoffs
• FANOK: knockoffs in linear time
• Gene hunting with hidden Markov model knockoffs
• A Scale-Free Approach for False Discovery Rate Control in Generalized Linear Models
• Threshold Selection in Feature Screening for Error Rate Control
• Controlling False Discovery Rate Using Gaussian Mirrors
• False Discovery Rate Control via Data Splitting
• A generalized knockoff procedure for FDR control in structural change detection
• Adaptive novelty detection with false discovery rate guarantee
• Generating knockoffs via conditional independence
• Deep latent variable models for generating knockoffs
• CoxKnockoff: controlled feature selection for the Cox model using knockoffs
• Local false discovery rate estimation with competition-based procedures for variable selection
• Controlling the false discovery rate by a latent Gaussian copula knockoff procedure
• FDR control and power analysis for high-dimensional logistic regression via Stabkoff
• False discovery rate-controlled multiple testing for union null hypotheses: a knockoff-based approach
• Support recovery of Gaussian graphical model with false discovery rate control
• Reconciling model-X and doubly robust approaches to conditional independence testing
• Overview of research advance for knockoff methods
• Sequential knockoffs for continuous and categorical predictors: with application to a large psoriatic arthritis clinical trial pool
• Stab-GKnock: controlled variable selection for partially linear models using generalized knockoffs
• An adaptive threshold for outlier detection in high-dimensional settings
• Joint mirror procedure: controlling false discovery rate for identifying simultaneous signals
• Derandomized Truncated D-vine Copula Knockoffs with e-values to control the false discovery rate
• High-Dimensional Knockoffs Inference for Time Series Data
• Ah-knockoff: false discovery rate control in high-dimensional additive hazards models
• ARK: robust knockoffs inference with coupling
• Model-Free Statistical Inference on High-Dimensional Data
• Variable selection in latent variable models via knockoffs: an application to international large-scale assessment in education
• Controlling false discovery rate for mediator selection in high-dimensional data
• Knockoffs for partially exchangeable categorical covariates
• A novel explainable artificial intelligence framework using knockoffs techniques with applications to sports analytics
• Approximate co-sufficient sampling with regularization
• Controlling FDR in selecting group-level simultaneous signals from multiple data sources with application to the National COVID Collaborative Cohort data
• Private Model-X knockoffs via Johnson-Lindenstrauss transform
• Controlling the false discovery rate via knockoffs

Describes a project that uses

Uses Software

• glasso

This page was built for publication: Robust inference with knockoffs

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2196226)