Semi-supervised multiple testing

Mary, David; Roquain, Étienne

doi:10.1214/22-ejs2050

Cited by 9 publications

(8 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Remark 9. A similar martingale argument was developed by Mary and Roquain [41] to prove that the BH procedure (λ = 1) applied to the RANC p-values controls the FDR. A main distinction is that they require partial exchangeability of the test statistics (condition (ii) in Theorem 3.1), which is weaker than the independence condition (b) in Theorem 4.1 but does not allow the case of uniformly stochastic dominance in condition (a).…”

Section: The Empirical Process Perspectivementioning

confidence: 91%

“…Specifically, our first method (RANC p-value) is the same as the conformal p-values to test for outliers proposed by Bates et al [3]. Our second method (empirical estimation of FDR) is essentially the same as the so-called "semi-supervised" multiple testing proposed by Mary and Roquain [41], and is closely related to the analysis of knockoff filters in Weinstein, Barber, and Candes [63]; they are all essentially derived from the martingale argument in Storey, Taylor, and Siegmund [56]. Our third method (estimator of local-FDR) is closely related to the method in Soloff, Xiang, and Fithian [53] which assumes a known null distribution.…”

Section: Overview Of the Proposed Methodsmentioning

confidence: 99%

“…See Marandon et al [39] and Liang, Sesia, and Sun [35] for some extensions. Another closely related perspective is to view multiple testing with negative controls as a semi-supervised problem [41], a term borrowed from the machine learning literature. Motivated by problems in astrostatistics, Mary and Roquain [41] proposed to use the same p-value using the empirical null viewpoint.…”

Section: Literature Reviewmentioning

confidence: 99%

“…Another closely related perspective is to view multiple testing with negative controls as a semi-supervised problem [41], a term borrowed from the machine learning literature. Motivated by problems in astrostatistics, Mary and Roquain [41] proposed to use the same p-value using the empirical null viewpoint. They analyzed the BH procedure applied to such p-values using a martingale argument and provided some further optimality results.…”

Section: Literature Reviewmentioning

confidence: 99%

“…4 The uniformly stochastic dominance condition (a) arises naturally in our proof of the first inequality. It remains unclear to us whether this can be allowed in the proof in Mary and Roquain [41], as their martingale is not indexed by the rejection threshold. When the marginal distributions of the null test statistics and internal negative controls are the same, the gap…”

Section: The Empirical Process Perspectivementioning

confidence: 99%

See 4 more Smart Citations

Simultaneous Hypothesis Testing Using Internal Negative Controls with An Application to Proteomics

Zhao¹

2023

Preprint

View full text Add to dashboard Cite

Negative control is a common technique in scientific investigations and broadly refers to the situation where a null effect ("negative result") is expected. Motivated by a real proteomic dataset, we will present three promising and closely connected methods of using negative controls to assist simultaneous hypothesis testing. The first method uses negative controls to construct a permutation p-value for every hypothesis under investigation, and we give several sufficient conditions for such p-values to be valid and positive regression dependent on the set (PRDS) of true nulls. The second method uses negative controls to construct an estimate of the false discovery rate (FDR), and we give a sufficient condition under which the step-up procedure based on this estimate controls the FDR. The third method, derived from an existing ad hoc algorithm for proteomic analysis, uses negative controls to construct a nonparametric estimator of the local false discovery rate. We conclude with some practical suggestions and connections to some closely related methods that are propsed recently.

show abstract

Section: The Empirical Process Perspectivementioning

confidence: 91%

Section: Overview Of the Proposed Methodsmentioning

confidence: 99%

Section: Literature Reviewmentioning

confidence: 99%

Section: Literature Reviewmentioning

confidence: 99%

Section: The Empirical Process Perspectivementioning

confidence: 99%

See 3 more Smart Citations

Simultaneous Hypothesis Testing Using Internal Negative Controls with An Application to Proteomics

Zhao¹

2023

Preprint

View full text Add to dashboard Cite

show abstract

Conformal link prediction for false discovery rate control

Marandon

2024

TEST

View full text Add to dashboard Cite

Most link prediction methods return estimates of the connection probability of missing edges in a graph. Such output can be used to rank the missing edges from most to least likely to be a true edge, but does not directly provide a classification into true and nonexistent. In this work, we consider the problem of identifying a set of true edges with a control of the false discovery rate (FDR). We propose a novel method based on high-level ideas from the literature on conformal inference. The graph structure induces intricate dependence in the data, which we carefully take into account, as this makes the setup different from the usual setup in conformal inference, where data exchangeability is assumed. The FDR control is empirically demonstrated for both simulated and real data.

show abstract

The MUSEHubbleUltra Deep Field surveys: Data release II

Bacon

Brinchmann

Conseil

et al. 2023

A&A

View full text Add to dashboard Cite

We present the second data release of the MUSE Hubble Ultra-Deep Field surveys, which includes the deepest spectroscopic survey ever performed. The MUSE data, with their 3D content, amazing depth, wide spectral range, and excellent spatial and medium spectral resolution, are rich in information. Their location in the Hubble ultra-deep field area, which benefits from an exquisite collection of ancillary panchromatic information, is a major asset. This update of the first release incorporates a new 141-hour adaptive-optics-assisted MUSE eXtremely Deep Field (MXDF; 1 arcmin diameter field of view) in addition to the reprocessed 10-hour mosaic (3 × 3 arcmin 2 ) and the single 31-hour deep field (1 × 1 arcmin 2 ). All three data sets were processed and analyzed homogeneously using advanced data reduction and analysis methods. The 3σ point-source flux limit of an unresolved emission line reaches 3.1×10 −19 and 6.3×10 −20 erg s −1 cm −2 at 10-and 141-hour depths, respectively. We have securely identified and measured the redshift of 2221 sources, an increase of 41% compared to the first release. With the exception of eight stars, the collected sample consists of 25 nearby galaxies (z < 0.25), 677 [O ii] emitters (z = 0.25 − 1.5), 201 galaxies in the MUSE redshift desert range (z = 1.5 − 2.8), and 1308 Lyα emitters (z = 2.8 − 6.7). This represents an order of magnitude more redshifts than the collection of all spectroscopic redshifts obtained before MUSE in the Hubble ultra-deep field area (i.e., 2221 versus 292). At high redshift (z > 3), the difference is even more striking, with a factor of 65 increase (1308 versus 20). We compared the measured redshifts against three published photometric redshift catalogs and find the photo-z accuracy to be lower than the constraints provided by photo-z fitting codes. Eighty percent of the galaxies in our final catalog have an HST counterpart. These galaxies are on average faint, with a median AB F775W magnitude of 25.7 and 28.7 for the [O ii] and Lyα emitters, respectively. Fits of their spectral energy distribution show that these galaxies tend to be low-mass star-forming galaxies, with a median stellar mass of 6.2 × 10 8 M and a median star-formation rate of 0.4 M yr −1 . We measured the completeness of our catalog with respect to HST and found that, in the deepest 141-hour area, 50% completeness is achieved for an AB magnitude of 27.6 and 28.7 (F775W) at z = 0.8 − 1.6 and z = 3.2 − 4.5, respectively. Twenty percent of our catalog, or 424 galaxies, have no HST counterpart. The vast majority of these new sources are high equivalent-width z > 2.8 Lyα emitters that are detected by MUSE thanks to their bright and asymmetric broad Lyα line. We release advanced data products, specific software, and a web interface to select and download data sets.

show abstract

Semi-supervised multiple testing

Cited by 9 publications

References 51 publications

Simultaneous Hypothesis Testing Using Internal Negative Controls with An Application to Proteomics

Simultaneous Hypothesis Testing Using Internal Negative Controls with An Application to Proteomics

Conformal link prediction for false discovery rate control

The MUSEHubbleUltra Deep Field surveys: Data release II

Contact Info

Product

Resources

About