Sample Splitting Research Articles

False Discovery Rate Control Under General Dependence By Symmetrized Data Aggregation

We develop a new class of distribution-free multiple testing rules for false discovery rate (FDR) control under general dependence. A key element in our proposal is a symmetrized data aggregation (SDA) approach to incorporating the dependence structure via sample splitting, data screening, and information pooling. The proposed SDA filter first constructs a sequence of ranking statistics that fulfill global symmetry properties, and then chooses a data-driven threshold along the ranking to control the FDR. The SDA filter substantially outperforms the Knockoff method in power under moderate to strong dependence, and is more robust than existing methods based on asymptotic p-values. We first develop finite-sample theories to provide an upper bound for the actual FDR under general dependence, and then establish the asymptotic validity of SDA for both the FDR and false discovery proportion control under mild regularity conditions. The procedure is implemented in the R package sdafilter. Numerical results confirm the effectiveness and robustness of SDA in FDR control and show that it achieves substantial power gain over existing methods in many settings.

Challenges in Obtaining Valid Causal Effect Estimates with Machine Learning Algorithms.

Unlike parametric regression, machine learning (ML) methods do not generally require precise knowledge of the true data generating mechanisms. As such, numerous authors have advocated for ML methods to estimate causal effects. Unfortunately, ML algorithmscan perform worse than parametric regression. We demonstrate the performance of ML-based single- and double-robust estimators. We use 100 Monte Carlo samples with sample sizes of 200, 1200, and 5000 to investigate bias and confidence interval coverage under several scenarios. In a simple confounding scenario, confounders were related to the treatment and the outcome via parametric models. In a complex confounding scenario, the simple confounders were transformed to induce complicated nonlinear relationships. In the simple scenario, when ML algorithms were used, double-robust estimators were superior to single-robust estimators. In the complex scenario, single-robust estimators with ML algorithms were at least as biased as estimators using misspecified parametric models. Double-robust estimators were less biased, but coverage was well below nominal. The use of sample splitting, inclusion of confounder interactions, reliance on a richly specified ML algorithm, and use of doubly robust estimators was the only explored approach that yielded negligible bias and nominal coverage. Our results suggest that ML based singly robust methods should be avoided.

Changes in profile distribution and chemical properties of natural nanoparticles in paddy soils as affected by long-term rice cultivation

Systematic studies on the genesis, properties, and distribution of natural nanoparticles (NNPs) in soil remain scarce. This study examined a soil chronosequence of continuous paddy field land use for periods ranging from 0 to 1 000 years to determine how NNPs in soil changed at the early stages of soil genesis in eastern China. Soil samples were collected from coastal reclaimed paddy fields that were cultivated for 0, 50, 100, 300, 700, and 1 000 years. Natural nanoparticles were isolated and characterized along with bulk soil samples (< 2-mm fraction) for selected physical and chemical properties. The NNP content increased with increasing soil cultivation age at 60 g m−2 year−1, which was related to decreasing soil electrical conductivity (172–1 297 µS cm−1) and NNP zeta potentials (from –22 to –36 mV) with increasing soil cultivation age. Changes in several NNP properties, such as pedogenic iron oxide and total organic carbon contents, were consistent with those of the bulk soils across the soil chronosequence. Notably, changes in NNP iron oxide content were obvious and illustrated active chemical weathering, pedogenesis, and potential impacts on the microbial community. Redundancy analysis demonstrated that the soil cultivation age was the most important factor affecting NNP properties, contributing 60.7% of the total variation. Cluster and principal component analysis (PCA) revealed splitting of NNP samples into age groups of 50–300 and 700–1 000 years, indicating rapid evolution of NNP properties, after an initial period of desalinization (approximately 50 years). Overall, this study provides new insights into NNP evolution in soil during pedogenesis and predicting their influences on agriculture and ecological risks over millennial-scale rice cultivation.

Meat species identification accuracy improvement using sample set portioning based on joint x-y distance and laser-induced breakdown spectroscopy.

Laser-induced breakdown spectroscopy (LIBS) was suitable for the identification of meat species due to fast and less sample preparation. However, the problem of low accuracy rate of the recognition model caused by improper selection of training set samples by random split has severely restricted the development of LIBS in meat detection. Sample set portioning based on the joint x-y distance (SPXY) method was applied for dividing the meat spectra into a training set and a test set. Then, the five kinds of meat samples (shrimp, chicken, beef, scallop, and pig liver) were classified by the support vector machine (SVM). With the random split method, Kennard-Stone method, and SPXY method, the recognition accuracies of the SVM model were 90.44%, 91.95%, and 94.35%, respectively. The multidimensional scaling method was used to visualize the results of the sample split for the interpretation of the classification. The results showed that the identification performance of the SPXY method combined with the SVM model was best, and the accuracy rates of shrimp, chicken, beef, scallop, and pig liver were 100.00%, 100.00%, 100.00%, 78.57%, and 92.00%, respectively. Moreover, to verify the broad adaptability of the SPXY method, the linear discriminant analysis model, the K-nearest neighbor model, and the ensemble learning model were applied as the meat species identification model. The results demonstrated that the accuracy rate of the classification model can be improved with the SPXY method. In light of the findings, the proposed sample portioning method can improve the accuracy rate of the recognition model using LIBS.

Abstract LB059: Dual methylation and mutation analysis from a single input of plasma cfDNA without sample splitting

Abstract Genomic and epigenetic changes are known to be the underlying mechanism of cancer development. Assessing the genomic (mutation) and epigenetic (methylation) changes has become increasingly important for cancer detection and treatment selection, especially in the field of cfDNA-based liquid biopsy. Targeted Next-Generation Sequencing (NGS) enables in-depth analysis of specific genomic regions that is essential for the clinical application of liquid biopsy. However, current targeted NGS assays are complicated, lengthy, and inefficient. Two major hurdles of existing targeted NGS methods make them impractical for the clinical setting. First, there is no comprehensive, end to end, kit solution available for targeted methylation sequencing (TMS) from sample to sequencing ready library. Second, the low yield of cfDNA from clinical blood samples presents a major challenge for multi-modal analyses on one sample. Thus, an assay that is capable of both methylation and mutation analysis would be optimal for clinical research and future diagnostic assays.Here, we report the performance of Point-n-Seq, a kit solution that can provide in-depth DNA analysis with highly flexible and customizable focused panels to enable both methylation and mutation analysis without sample splitting. Custom panels of 10 to 1000s of markers can be designed with &amp;gt;99% first-pass success rate. We conducted both performance validation and multi-center, multi-operator, reproducibility studies. With spike-in titration of cancer cell-line gDNA with known mutation and methylation profile, we achieved a reliable detection level down to 0.003% of tumor DNA with a linear relationship between the measured and expected fractions.In a pilot study, we designed a colorectal cancer (CRC) TMS panel covering 560 methylation markers and a mutation panel with &amp;gt;350 hotspot mutations in 22 genes. From only 1ml of plasma from late stage CRC patients, we detected cancer-specific methylation signals in all samples tested, and hotspot mutations with clinical implications. Next, a tumor-informed mutation panel covering approximately 100 personalized mutations for each patient was compared to the tumor-independent CRC methylation panel to analyze 23 stage I and II CRC patient plasma samples. The initial results showed that tumor-independent TMS assay achieved a comparable or superior limit of detection compared to the personalized tumor-informed approach. Moreover, to further improve assay sensitivity, metrics for cfDNA size can be integrated into the analysis of the same workflow without sample splitting. The Point-n-Seq kit, with a highly streamlined workflow and dual methylation and mutation analysis, provides a highly sensitive and robust non-invasive method for cancer detection, monitoring and treatment guidance. Citation Format: Grace Q. Zhao, Yun Bao, Heng Wang, Vanita S. Natu, Jayashree Joshi, Yingmin Wang, Elena Arzumanyan, Wanping Wang, Jianmin Wang, John Coller, Shengrong Lin. Dual methylation and mutation analysis from a single input of plasma cfDNA without sample splitting [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr LB059.

Compulsive and compensative buying among online shoppers: An empirical study.

The present study examines the phenomenon of compensative and compulsive buying among online shoppers. Firstly, the obtained empirical data make it possible to estimate the prevalence of compensative and compulsive buying among the general population of Poles aged 15 years old and over, with the sample split into users and non-users of the e-commerce market offer. Secondly, the conducted analysis shows to what extent the prevalence of compulsive and compensative buying is differentiated by the frequency of online shopping, by the extent of the expenditures on online shopping compared with offline shopping, by attitudes towards online shopping, and by sociodemographic conditions (gender, age, monthly net income of household). The findings come from a survey conducted in 2019 based on a nationwide statistically representative sample of 1,000 Poles aged 15 years old and over. Drawing on this survey based on the German Compulsive Buying Indicator (GCBI), the prevalence of compulsive buying is observed at about 3% and compensative buying at about 12%. Dividing the general population into online and offline shoppers, one can see serious differences between both target groups; the share of compulsive and compensative buyers in the segment of online shoppers amounts to 3.6% and 16.9%, while among non-online shoppers- 3.3% and 10.1%. The strongest susceptibility to compulsive buying is characteristic of female online shoppers having very positive attitudes towards online shopping and doing online shopping very frequently.

A dual target sequencing solution to assess genomic and epigenomic alterations in cell-free DNA with no sample splitting.

3043 Background: Assessing the genomic and epigenomic changes on plasma cell-free DNA (cfDNA) using next-generation sequencing (NGS) has become increasingly important for cancer detection and treatment selection guidance. However, two major hurdles of existing targeted NGS methods make them impractical for the clinical setting. First, there is no comprehensive, end to end, kit solution available for targeted methylation sequencing (TMS), let alone one that analyzes both mutation and methylation information in one assay. Second, the low yield of cfDNA from clinical blood samples presents a major challenge for conducting multi-omic analysis. Thus, an assay that is capable of both genomic and epigenomic analysis would be advantageous for clinical research and future diagnostic assays. Methods: Here, we report the performance of Point-n-SeqTM dual analysis, a kit solution that can provide in-depth DNA analysis with highly flexible and customizable focused panels to enable both genomic and epigenomic analysis without sample splitting. With custom panels of tens to thousands of markers designed with &gt; 99% first-pass success rate, we conducted both performance validation and multi-center, multi-operator, reproducibility studies. Using spike-in titration of cancer cell-line gDNA with known mutation and methylation profiles, Point-n-Seq assay achieved a reliable detection level down to 0.003% of tumor DNA with a linear relationship between the measured and expected fractions. Benchmarked with conventional targeted sequencing and methylation sequencing, Point-n-Seq solution also demonstrated improved performance, speed and shortened hands-on time. Results: In a pilot clinical study, a colorectal cancer (CRC) TMS panel covering 560 methylation markers and a mutation panel with &gt; 350 hotspot mutations in 22 genes were used in the dual assay. Using 1ml of plasma from late-stage CRC patients, cancer-specific methylation signals were detected in all samples tested, and oncogenic mutations. In an early-stage cohort (33 stage I/II CRC patient ), comparison of the analysis between tumor-informed, personalized-mutation panels (̃100 private SNVs) for each patient and the tumor-independent CRC methylation panels were conducted. The initial results showed that tumor-independent TMS assay achieved a comparable detection compared to the personalized tumor-informed approach. Moreover, cfDNA size information (fragmentome) is also integrated into the analysis of the same Point-n-Seq workflow to improve the assay sensitivity. Conclusions: Point-n-Seq dual analysis is poised to advance both research and clinical applications of early cancer detection, minimal residual disease (MRD), and monitoring.

Measuring corporate tax rate and tax base avoidance of U.S. Domestic and U.S. multinational firms

We develop an approach based on publicly available data to decompose and quantify tax avoidance into two separate components: tax rate avoidance and tax base avoidance. Our measures are based on the average statutory tax rate, which accounts for the statutory tax rates across all transactions of a firm. We illustrate and validate our measures using simulation data, the Tax Reform Act of 1986, the Tax Cuts and Jobs Act of 2017, changes in tax rate avoidance and tax base avoidance across time, bonus depreciation time periods, several sample splits of U.S. multinational and domestic firms, differences across industries, and firms operating in tax haven locations. The measures allow regulators and researchers to gain insights into these two conceptually different tax avoidance strategies.

Boosting heritability: estimating the genetic component of phenotypic variation with multiple sample splitting

BackgroundHeritability is a central measure in genetics quantifying how much of the variability observed in a trait is attributable to genetic differences. Existing methods for estimating heritability are most often based on random-effect models, typically for computational reasons. The alternative of using a fixed-effect model has received much more limited attention in the literature.ResultsIn this paper, we propose a generic strategy for heritability inference, termed as “boosting heritability”, by combining the advantageous features of different recent methods to produce an estimate of the heritability with a high-dimensional linear model. Boosting heritability uses in particular a multiple sample splitting strategy which leads in general to a stable and accurate estimate. We use both simulated data and real antibiotic resistance data from a major human pathogen, Sptreptococcus pneumoniae, to demonstrate the attractive features of our inference strategy.ConclusionsBoosting is shown to offer a reliable and practically useful tool for inference about heritability.

Projection‐based and cross‐validated estimation in high‐dimensional Cox model

AbstractWe propose a projection‐based cross‐validation method for estimating a low‐dimensional parameter in the presence of a high‐dimensional nuisance parameter in the Cox regression model. We show that the proposed estimator is asymptotically normal, which enables us to conduct hypothesis test for the parameter of interest with high‐dimensional nuisance parameters. Three decision rules are presented to avoid the influence of random splitting of samples. Simulation studies indicate that our method is more powerful than that of Fang et al. (2017, JRSSB) when the coefficients of predictors are high‐dimensional and not very sparse. As an illustrative example, we apply our procedure to a breast cancer study.

Nonconvex Matrix Factorization From Rank-One Measurements

We consider the problem of recovering low-rank matrices from random rank-one measurements, which spans numerous applications including covariance sketching, phase retrieval, quantum state tomography, and learning shallow polynomial neural networks, among others. Our approach is to directly estimate the low-rank factor by minimizing a nonconvex least-squares loss function via vanilla gradient descent, following a tailored spectral initialization. When the true rank is bounded by a constant, this algorithm is guaranteed to converge to the ground truth (up to global ambiguity) with near-optimal sample complexity and computational complexity. To the best of our knowledge, this is the first guarantee that achieves near-optimality in both metrics. In particular, the key enabler of near-optimal computational guarantees is an implicit regularization phenomenon: without explicit regularization, both spectral initialization and the gradient descent iterates automatically stay within a region incoherent with the measurement vectors. This feature allows one to employ much more aggressive step sizes compared with the ones suggested in prior literature, without the need of sample splitting.

Re–examining Bhagwati hypothesis: the case of some selected countries in Sub-Saharan Africa

Bhagwati hypothesis opines that the overall impact of foreign direct investment (FDI) on economic growth is conditioned on countries’ level of integration with the international market. We test this hypothesis for some selected countries in sub-Saharan Africa (SSA). Does this hypothesis hold given our sample evidence? Yes! No! Maybe! We explain why. By invoking the sample splitting and threshold estimation technique, we find that the two measures of openness (trade openness and exports) mediate the FDI-economic growth relationship in three countries and this is an indication of complete Bhagwati hypothesis in these countries. Also, we find that, given the measure of openness, four countries exhibit incomplete Bhagwati hypothesis. Finally, we find no support for the Bhagwati hypothesis for most countries. Based on these findings, we argue that the validity of the Bhagwati hypothesis may be contingent on both country characteristics and the indicator of openness.

Charge carrier confinement in hematite and hematite / MoTe2 nanocomposites; the role of interface states

We report on the charge carrier confinement of the hematite and Hematite/Molybdenum di-telluride (α-Fe2O3@MoTe2) nanocomposites with special focus on the formation and role of interface states. X-ray diffraction (XRD) data confirms the existence of α-Fe2O3 in trigonally distorted rhombohedral while the MoTe2 in 2H hexagonal phase. Interfacial behavior and local chemical structure of the α-Fe2O3 and α-Fe2O3@MoTe2 nanocomposites were observed from the Mössbauer spectroscopy data. The hyperfine field splitting of the composite sample confirms the existence of the Fe3+ ions in two different octahedral environments, where 10% of surface Fe ions are in chemical interaction with the MoTe2 layers thereby making interfacial states. UV–Vis diffuse reflectance spectroscopy was employed to determine the band gap and band tail states, which are the key parameters in controlling the localization behavior. The conductivity, relaxation behavior and charge carrier confinement were observed by employing impedance spectroscopy.

Is the monetary policy behaviour of the South African Reserve Bank nonlinear

In spite of the apparent asymmetry in monetary policy behaviour, monetary policy rule exposition continues to be carried out in the linear context. The burgeoning literature on nonlinear monetary policy rules fail to capture nonlinearity appropriately, putting the precision and accuracy of such estimates in doubt for policy purposes. With the aid of sample splitting and threshold estimation technique, the current study considers the monetary policy behaviour and responses of the South African Reserve Bank (SARB) in the Taylor rule context. The chosen technique delivers a threshold estimate after 5,000 bootstrap replications with confidence intervals based on asymptotic theory devoid of nuisance parameters. The study finds nonlinear Taylor rule to adequately capture the policy behaviour of the SARB with threshold inflation of 5.2% and asymmetric responses below and above this threshold.

Is the monetary policy behaviour of the South African Reserve Bank nonlinear

In spite of the apparent asymmetry in monetary policy behaviour, monetary policy rule exposition continues to be carried out in the linear context. The burgeoning literature on nonlinear monetary policy rules fail to capture nonlinearity appropriately, putting the precision and accuracy of such estimates in doubt for policy purposes. With the aid of sample splitting and threshold estimation technique, the current study considers the monetary policy behaviour and responses of the South African Reserve Bank (SARB) in the Taylor rule context. The chosen technique delivers a threshold estimate after 5,000 bootstrap replications with confidence intervals based on asymptotic theory devoid of nuisance parameters. The study finds nonlinear Taylor rule to adequately capture the policy behaviour of the SARB with threshold inflation of 5.2% and asymmetric responses below and above this threshold.

Bearing defects detection based on standardized sample split

表面缺陷对轴承的性能和寿命存在严重影响。近年来，深度学习在缺陷检测中发挥了重要的作用，然而对于轴承检测而言，缺陷样本的采集耗时耗力。选择轴承内径作为研究对象，根据轴承的对称性特性提出一种规范化样本拆分方法，可有效扩充轴承样本数据集。分别采用不同的样本处理方法，而后利用ResNet网络训练轴承缺陷检测模型，进行多组对比实验，实验结果表明：直接采用原始图像进行网络训练，检测效果较差，模型的AUC (area under the curve)仅为0.558 0；对原始图像进行样本拆分，训练出的模型检测效果有所提升，其模型AUC提升为0.632 6；将原始图像进行4点透视变换校正后再进行网络训练，检测效果同样有所提升，其模型AUC提升为0.661 3；将原始图像进行透视变换校正且规范化样本拆分后，检测效果最好，其模型AUC增加为0.849 6。

Investment–cash flow sensitivity around the crisis: are African firms different?

PurposeThis paper examines the contentious relationship between investment and cash flow using the 2008–2009 credit supply shock as a form of the quasi-natural experiment.Design/methodology/approachPanel threshold models with unknown sample separation are estimated for a sample of publicly listed firms from nine African countries over the period 2003–2012. Using this approach reduces subjective or ex ante sample-splitting bias that is not accounted for in the extant literature.FindingsThe findings of the study indicate that investment–cash flow sensitivity is decreasing even during the global financial crisis (GFC) and for firms more likely to be financially constrained. The authors conclude that the usefulness of investment–cash flow sensitivity as a proxy for financial constraints is diminishing over time, even after directly addressing biases from ex ante subjective sample splitting and various forms of endogeneity.Originality/valueThe authors provide new empirical evidence from sharper tests of financial constraints for understudied African firms and highlight the need to relook at the usefulness of investment–cash flow sensitivity as a proxy of financial constraints.

ATRT-33. ENABLING RAPID CLASSIFICATION OF ATRT WITH NANOSTRING NCOUNTER PLATFORM

In recent years, using gene expression and methylation array platform, multiple research groups have reported the presence of at least three major Atypical Teratoid Rhabdoid Tumor (ATRT) subtypes that exhibit distinct epigenetic, transcriptomic and clinical features. Yet, utilizing ATRT subtypes in a clinical setting remains challenging due to a lack of suitable biological markers, limited sample quantities and relatively high cost of current assays. To address this gap between research and clinical practice, we have designed an assay that utilizes a custom 35 signature genes panel for the NanoString nCounter System and have created a flexible machine learning classifier package for ATRT tumour subtyping. We have analyzed 71 ATRT primary tumours with matching gene expression data using the 35 genes panel. 60% of the data was used for models training (10 repeats of 10-fold cross validation with subgroup balanced sample splitting) resulting in overall 94.6% training accuracy. The remaining 40% of the samples were used for model validation and the assay was able to achieve 92–100% accuracy with no subgroup bias. To demonstrate the flexibility of the workflow, we have tested it against other transcriptome-based methods such as gene expression array and RNASeq. We have also demonstrated its use in samples that were not classifiable by methylation-based method. We are presenting here a rapid and accurate ATRT subtyping assay for clinical usage that is compatible with archived ATRT tissues.

Migration enclave and downward economic growth in Africa

PurposeThe African region has experienced relatively lower economic growth and higher outflow of migration over the years. The purpose of this research paper, therefore, is to focus on the African region to investigate whether or not there is any link between the poor economic growth and rising outflow of migration.Design/methodology/approachEmpirical data spanning from 1990 to 2015 are collected from reliable sources for 41 countries belonging to the African region. Appropriate estimating methodology that controls for unobserved heterogeneity both across time and across countries, and endogeneity is employed.FindingsThe results revealed that the migration has adversely influenced the economic growth of the African region as a whole. The splitting of sample into male and female migration also reflected the fact that the unsatisfied economic growth of the African region could be explained by the ever rising migration level. Other determinants such as employment and growth of physical capital have helped the region in the growth journey. Human capital has not played a vital role in economic growth as it is adversely affected by migration. Further, the study found support for the positive impact of moderate inflation on economic growth. The obtained results are robust to alternative methodologies and hence would be beneficial for policymakers.Originality/valueThe present study provides for the first time comprehensive empirical evidence on the relationship between migration and economic growth by focusing on Africa. Therefore, this study would be of prime importance for policymakers.

Minimax Estimation of Divergences Between Discrete Distributions

We study the minimax estimation of $\alpha$-divergences between discrete distributions for integer $\alpha\ge 1$, which include the Kullback--Leibler divergence and the $\chi^2$-divergences as special examples. Dropping the usual theoretical tricks to acquire independence, we construct the first minimax rate-optimal estimator which does not require any Poissonization, sample splitting, or explicit construction of approximating polynomials. The estimator uses a hybrid approach which solves a problem-independent linear program based on moment matching in the non-smooth regime, and applies a problem-dependent bias-corrected plug-in estimator in the smooth regime, with a soft decision boundary between these regimes.