Data-splitting Approach Research Articles

Concentration division for adsorption coefficient prediction using machine learning with Abraham descriptors: Data-splitting approach comparison and critical factors identification

Machine learning (ML) including Abraham descriptors from polyparameter linear free energy relationships (pp-LFERs) has been a popular method for the adsorption coefficient (Kd) prediction. However, Abraham descriptors from pp-LFERs are concentration-dependent and the significance of these descriptors can change over different adsorbate concentrations. Ignoring concentration effects on the adsorption process and Kd prediction will hinder the understanding of interactions among solutes, solvents and adsorbents at different equilibrium concentration (Ce) ranges. Therefore, our study first systematically investigated the concentration effects on micropollutant adsorption to carbon-based adsorbents using ML with Abraham descriptors. Concentration-selection approach, as a new data-splitting approach, divided the whole dataset according to the different Ce ranges. This concentration-selection approach performed better than the data-splitting approach used in previous studies. After the ML models were built in different Ce subsets, Shapley values were calculated to quantify input descriptor contributions. The results indicated specific surface area (BET) was the only critical factor when Ce was in the highest range. The importance of Abraham descriptors increased gradually when Ce decreased. Total pore volume (Vt) was a far less important feature than BET for Kd prediction. Critical factors identified at different Ce ranges for Kd prediction provide a guidance for novel carbon-based adsorbent design.

MP35-14 A NOVEL PREOPERATIVE SCORING SYSTEM FOR PREDICTING SUCCESSFUL RENAL ACCESS FOR ULTRASOUND-GUIDED PERCUTANEOUS NEPHROLITHOTOMY

You have accessJournal of UrologyCME1 Apr 2023MP35-14 A NOVEL PREOPERATIVE SCORING SYSTEM FOR PREDICTING SUCCESSFUL RENAL ACCESS FOR ULTRASOUND-GUIDED PERCUTANEOUS NEPHROLITHOTOMY Ukrit Rompsaithong, Heiko Yang, Wilson Sui, Rei Unno, David Bayne, Justin Ahn, Marshall Stoller, and Thomas Chi Ukrit RompsaithongUkrit Rompsaithong More articles by this author , Heiko YangHeiko Yang More articles by this author , Wilson SuiWilson Sui More articles by this author , Rei UnnoRei Unno More articles by this author , David BayneDavid Bayne More articles by this author , Justin AhnJustin Ahn More articles by this author , Marshall StollerMarshall Stoller More articles by this author , and Thomas ChiThomas Chi More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000003269.14AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Renal access remains a challenging but important step in successful percutaneous nephrolithotomy (PCNL). Understanding factors that predict successful access will help operative planning. With ultrasound guidance for renal access being adopted for widespread clinical practice, this study aimed to develop and validate a novel preoperative scoring system for predicting renal access difficulty for ultrasound-guided PCNL. METHODS: We analyzed consecutive patients undergoing ultrasound-guided PCNL from the Registry for Stones of the Kidney and Ureter (ReSKU) between 2015 and 2022. A risk model was created using a randomized data-splitting approach, using 70% for model development and 30% for model validation. Ultrasound-guided access failure was defined as the inability to puncture the renal calyx of choice using ultrasound-only guidance. To create a multivariate model, the preoperative variables were determined using forward selection and modified according to clinical judgment. The predictive scores were created by weighting the coefficients of each significant factor. The total scores were classified into three risk groups according to the prevalence and the correlation of difficult renal access. The final model was validated. RESULTS: Three hundred renal units were used for model development and 90 for validation. Four significant factors were identified in the final model and scored: body mass index ≥40 kg/m2 (2 points), no/mild hydronephrosis (4 points), staghorn (5 points), and diverticular stone (7 points) (Table 1). The total scores were categorized into low (0-4 points), moderate (5-8 points), and high difficulty (9-18 points), with a failure rate of 4.1%, 15.4%, and 50% in each group, respectively. This scoring system had a sensitivity of 78.1% and specificity of 80.7% in predicting failures over the moderate difficulty cut-point. The area under the receiver-operating characteristic (ROC) curve in the final model and the validation group were 85.5% and 87.7%, respectively. CONCLUSIONS: Our novel scoring system may be an effective screening tool to identify difficult cases for ultrasound-guided access. In these patients, we suggest having backup imaging modalities available for alternate means of achieving access. Source of Funding: None © 2023 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 209Issue Supplement 4April 2023Page: e474 Advertisement Copyright & Permissions© 2023 by American Urological Association Education and Research, Inc.MetricsAuthor Information Ukrit Rompsaithong More articles by this author Heiko Yang More articles by this author Wilson Sui More articles by this author Rei Unno More articles by this author David Bayne More articles by this author Justin Ahn More articles by this author Marshall Stoller More articles by this author Thomas Chi More articles by this author Expand All Advertisement PDF downloadLoading ...

Covariate Powered Cross-Weighted Multiple Testing

AbstractA fundamental task in the analysis of data sets with many variables is screening for associations. This can be cast as a multiple testing task, where the objective is achieving high detection power while controlling type I error. We consider m hypothesis tests represented by pairs ((Pi,Xi))1≤i≤m of p-values Pi and covariates Xi, such that Pi⊥Xi if Hi is null. Here, we show how to use information potentially available in the covariates about heterogeneities among hypotheses to increase power compared to conventional procedures that only use the Pi. To this end, we upgrade existing weighted multiple testing procedures through the independent hypothesis weighting (IHW) framework to use data-driven weights that are calculated as a function of the covariates. Finite sample guarantees, for example false discovery rate control, are derived from cross-weighting, a data-splitting approach that enables learning the weight-covariate function without overfitting as long as the hypotheses can be partitioned into independent folds, with arbitrary within-fold dependence. IHW has increased power compared to methods that do not use covariate information. A key implication of IHW is that hypothesis rejection in common multiple testing setups should not proceed according to the ranking of the p-values, but by an alternative ranking implied by the covariate-weighted p-values.

On statistical classification with incomplete covariates via filtering

This article deals with the problem of classification when some of the covariates may have missing parts. Here, it is allowed for both the training sample as well as the new unclassified observation to have missing parts in the covariates. In fact, it is shown in Remark 3.3 that in classification the reconstruction/imputation of the missing part of a new unclassified observation (which is to be classified) can be counter-productive in terms of the error rates. Furthermore, unlike many of the results in the literature, where covariate fragments are usually assumed to be missing completely at random, we do not impose such assumptions here. Given the observed parts of the covariates, we construct a kernel-type classifier which is straightforward to implement. The proposed classifier is constructed based on d-dim covariate vectors that are obtained from the original covariates (by moving from the space to ), where itself is a parameter that has to be estimated. To estimate various parameters, we employ an easy-to-implement data-splitting approach.

Comparing dominance hierarchy methods using a data-splitting approach with real-world data

Abstract The development of numerical methods for inferring social ranks has resulted in an overwhelming array of options to choose from. Previous work has established the validity of these methods through the use of simulated datasets, by determining whether a given ranking method can accurately reproduce the dominance hierarchy known to exist in the data. Here, we offer a complementary approach that assesses the reliability of calculated dominance hierarchies by asking whether the calculated rank order produced by a given method accurately predicts the outcome of a subsequent contest between two opponents. Our method uses a data-splitting “training–testing” approach, and we demonstrate its application to real-world data from wild vervet monkeys (Chlorocebus pygerythrus) collected over 3 years. We assessed the reliability of seven methods plus six analytical variants. In our study system, all 13 methods tested performed well at predicting future aggressive outcomes, despite some differences in the inferred rank order produced. When we split the dataset with a 6-month training period and a variable testing dataset, all methods predicted aggressive outcomes correctly for the subsequent 10 months. Beyond this 10-month cut-off, the reliability of predictions decreased, reflecting shifts in the demographic composition of the group. We also demonstrate how a data-splitting approach provides researchers not only with a means of determining the most reliable method for their dataset but also allows them to assess how rank reliability changes among age–sex classes in a social group, and so tailor their choice of method to the specific attributes of their study system.

Minimax Estimation of the Volume of a Set Under the Rolling Ball Condition

ABSTRACTWe consider the problem of estimating the volume of a compact domain in a Euclidean space based on a uniform sample from the domain. We assume that the domain has a boundary with positive reach. We propose a data-splitting approach to correct the bias of the plug-in estimator based on the sample α-convex hull. We show that this simple estimator achieves a minimax lower bound that we derive. Some numerical experiments corroborate our theoretical findings. Supplementary materials for this article are available online.

Non-Parametric Inference for Density Modes

Summary We derive non-parametric confidence intervals for the eigenvalues of the Hessian at modes of a density estimate. This provides information about the strength and shape of modes and can also be used as a significance test. We use a data-splitting approach in which potential modes are identified by using the first half of the data and inference is done with the second half of the data. To obtain valid confidence sets for the eigenvalues, we use a bootstrap based on an elementary symmetric polynomial transformation. This leads to valid bootstrap confidence sets regardless of any multiplicities in the eigenvalues. We also suggest a new method for bandwidth selection, namely choosing the bandwidth to maximize the number of significant modes. We show by example that this method works well. Even when the true distribution is singular, and hence does not have a density (in which case cross-validation chooses a zero bandwidth), our method chooses a reasonable bandwidth.

Selective correlations; not voodoo

The problem of “voodoo” correlations–exceptionally high observed correlations in selected regions of the brain–is well recognized in neuroimaging. It arises when quantities of interest are estimated from the same data that was used to select them as interesting. In statistical terminology, the problem of inference following selection from the same data is that of selective inference. Motivated by the unwelcome side-effects of splitting the data– the recommended remedy–we adapt the recent developments in selective inference in order to construct confidence intervals (CIs) with good reproducibility prospects, even if selection and estimation are done with the same data. These intervals control the expected proportion of non-covered correlations in the selected voxels—the False Coverage Rate (FCR). They extend further toward zero than standard intervals, thus attenuating the impression made by highly biased observed correlations. They do so adaptively, in that they coincide with the standard CIs when far away from the selection point. We complement existing analytic proofs with a simulation, showing that the proposed intervals control the FCR in realistic social neuroscience problems. We also suggest a “confidence calibration plot”, to allow the intervals to be reported in a clear and interpretable way. Applying the proposed methodology on a loss-aversion study, we demonstrate that with the sample size and selection type employed, selection bias is considerable. Finally, selective intervals are compared to the currently recommended data-splitting approach. We discover that our approach has more power and typically more informative, as no data is discarded.Computation of the intervals is implemented in an accompanying software package.

Operating Speed Prediction Models for Horizontal Curves on Rural Four-Lane Highways

Previous studies documented that a uniform design speed does not necessarily guarantee design consistency on rural two-lane facilities. Since a similar process is also followed for four-lane rural highways, it is reasonable to assume that similar inconsistencies could be found on such roadways. The operating speed-based method has been extensively used in other countries as the primary method to examine design consistency. Numerous studies have been completed on rural two-lane highways for predicting operating speeds and evaluating design consistency. However, few studies have considered these issues for rural four-lane highways. Therefore, prediction models for rural four-lane highways are needed. This study aims to develop models to predict operating speeds on horizontal curves of rural four-lane highways. A parallel study documented that speeds on inside and outside lanes are different; therefore, two multiple linear regression models are developed. For the inside lane, the significant factors are shoulder type, median type, pavement type, approaching section grade, and horizontal curve length. For the outside lane, factors include shoulder type, median type, approaching section grade, presence of approaching curve, and curve radius and length. The factors in the two models indicate that the curve itself mainly influences a driver's speed choice. The models were validated by using the data-splitting approach, and validation shows that there are no statistical differences between the predicted and field-observed operating speeds.

Development and validation of a model predicting graft survival after liver transplantation

This study aimed to develop and validate a comprehensive model that predicts survival after liver transplantation based on pretransplant donor and recipient characteristics. Complete data were available from the United Network for Organ Sharing for 20,301 persons who underwent liver transplantation in the United States between 1994 and 2003. Proportional-hazards regression was used to identify the donor and recipient characteristics that best predicted survival and incorporate these characteristics in a multivariate model. A data-splitting approach was used to compare survival predicted by the model to the observed survival in samples not used in the derivation of the model. A model was derived using 4 donor characteristics (age, cold ischemia time, gender, and race/ethnicity) and 9 recipient characteristics (age, body max index, model for end-stage liver disease score, United Network for Organ Sharing priority status, gender, race/ethnicity, diabetes mellitus, cause of liver disease, and serum albumin) that adequately predicted survival after liver transplantation in patients without hepatitis C virus, and a slightly different model was used for patients with hepatitis C virus. The models illustrate that variations in both pretransplant donor and recipient characteristics have a large effect on posttransplant survival. In conclusion, the models presented here can be used to derive scores that are proportional to the excess risk of graft loss after liver transplantation for potential donors, recipients, or donor/recipient combinations. The models may be used to inform liver transplant candidates and their doctors what posttransplant survival would be expected when a given donor is offered and may be particularly helpful for marginal or high-risk donors.

An iterated classification rule based on auxiliary pseudo-predictors

This article proposes a two-step iterative procedure to improve the misclassification error rate of an initial classification rule. The first step involves an iterative method for generating a sequence of classifiers from the initial one; this is based on the augmentation of the feature vector with some new pseudo-predictors. Unlike other components of the feature vector, these new pseudo-predictors tend to provide information primarily on the performance or correctness of the classifier itself. The second step of the proposed procedure “pools together” the classifiers constructed in step one in order to produce a new classifier which is far more effective (in an asymptotic sense) than the initial classifier. In addition to these results, a data-splitting approach for selecting the number of iterations will also be discussed. Both the mechanics and the asymptotic validity of the proposed procedure are studied.

New equations for estimating body cell mass from bioimpedance parallel models in healthy older Germans.

The objectives of this study were to assess for elderly Germans the validity of existing equations for predicting body cell mass (BCM) and to develop from single- and multifrequency bioimpedance (SFBIA, MFBIA) models new prediction equations. In a data-splitting approach, validation and cross-validation were performed in 160 healthy elderly (60- to 90-yr) subjects. BCM was determined using a tetrapolar bioimpedance analyzer (800 microA; 4 fixed frequencies: 1, 5, 50, and 100 kHz; electrodes placed to hand, wrist, ankle, and foot) and whole body (40)K counting as a reference method. New prediction equations were derived by multiple stepwise regression analysis. The Bland-Altman procedure was used for methods comparison. Relative to whole body counting, the manufacturer's equation overestimated BCM by 9% in men (P < 0.0001, paired t-test) and 4% in women (P = 0.002). Compared with the manufacturer's equation, the newly derived equations (r = 0.92, RMSE = 6-9%) improved accuracy (pure error = 13 vs. 7-8%) and reduced bias and limits of agreement. SFBIA and MFBIA equations did not differ in precision or accuracy. We conclude that the newly derived equations improved BCM estimates in the elderly compared with existing equations. There was no advantage of MFBIA over SFBIA equations.

A kernel-based combined classification rule

This article deals with a weighted-type combined classification rule where the combining is based on a data discretization and the “weights” are determined by exponential kernels. The smoothing parameter of the kernel is estimated by a data-splitting approach. Both the mechanics and the asymptotic validity of the proposed procedure are discussed.