Data-driven Selection Research Articles

Portable solid-state sensor for therapeutic monitoring of an antineoplastic drug; vinblastine in human plasma.

During cancer treatment, doses must be carefully administered and monitored to guarantee efficacy and minimize side-effects. A potentiometric sensor was developed for the direct real-time assay of a widely used antineoplastic drug (vinblastine (VB)) in plasma samples. Membrane cocktails were drop-casted over a glassy-carbon electrode coated with a lipophilic conducting polymer (polyaniline). The study investigated five cation exchangers, five plasticizers (of different polarities and dielectric constants), and four ionophores with different physicochemical characters on the sensor performance. The study substantiates a data-driven selection of the optimum membrane recipe. The latter included sodium tetraphenylborate as an ion exchanger, dioctylphthalate as a plasticizer, and hydroxypropyl-β-cyclodextrin as ionophore. The membrane proved a near-Nernstian slope of 37.5 mV per decade, a LOQ of 2.99 × 10−6 M, and a stable fast response. The selectivity study proved poor responses to common physiological ions. The developed sensor was used for the determination of VB in its pure powder form, marketed formulation, and plasma samples. The fast and direct sensor response enables a wide range of applications in quality control laboratories and clinical studies.

Choosing among Regularized Estimators in Empirical Economics: The Risk of Machine Learning

Abstract Many settings in empirical economics involve estimation of a large number of parameters. In such settings, methods that combine regularized estimation and data-driven choices of regularization parameters are useful. We provide guidance to applied researchers on the choice between regularized estimators and data-driven selection of regularization parameters. We characterize the risk and relative performance of regularized estimators as a function of the data-generating process and show that data-driven choices of regularization parameters yield estimators with risk uniformly close to the risk attained under the optimal (unfeasible) choice of regularization parameters. We illustrate using examples from empirical economics.

Adjustment for exploratory cut-off selection in randomized clinical trials with survival endpoint.

Defining the target population based on predictive biomarkers plays an important role during clinical development. After establishing a relationship between a biomarker candidate and response to treatment in exploratory phases, a subsequent confirmatory trial ideally involves only subjects with high potential of benefiting from the new compound. In order to identify those subjects in case of a continuous biomarker, a cut-off is needed. Usually, a cut-off is chosen that resulted in a subgroup with a large observed treatment effect in an exploratory trial. However, such a data-driven selection may lead to overoptimistic expectations for the subsequent confirmatory trial. Treatment effect estimates, probability of success, and posterior probabilities are useful measures for deciding whether or not to conduct a confirmatory trial enrolling the biomarker-defined population. These measures need to be adjusted for selection bias. We extend previously introduced Approximate Bayesian Computation techniques for adjustment of subgroup selection bias to a time-to-event setting with cut-off selection. Challenges in this setting are that treatment effects become time-dependent and that subsets are defined by the biomarker distribution. Simulation studies show that the proposed method provides adjusted statistical measures which are superior to naïve Maximum Likelihood estimators as well as simple shrinkage estimators.

Data-driven selection of conference speakers based on scientific impact to achieve gender parity.

A lack of diversity limits progression of science. Thus, there is an urgent demand in science and the wider community for approaches that increase diversity, including gender diversity. We developed a novel, data-driven approach to conference speaker selection that identifies potential speakers based on scientific impact metrics that are frequently used by researchers, hiring committees, and funding bodies, to convincingly demonstrate parity in the quality of peer-reviewed science between men and women. The approach enables high quality conference programs without gender disparity, as well as generating a positive spiral for increased diversity more broadly in STEM.

A data-driven selection of the number of clusters in the Dirichlet allocation model via Bayesian mixture modelling

ABSTRACTIn this paper, we consider a Bayesian mixture model that allows us to integrate out the weights of the mixture in order to obtain a procedure in which the number of clusters is an unknown quantity. To determine clusters and estimate parameters of interest, we develop an MCMC algorithm denominated by sequential data-driven allocation sampler. In this algorithm, a single observation has a non-null probability to create a new cluster and a set of observations may create a new cluster through the split-merge movements. The split-merge movements are developed using a sequential allocation procedure based in allocation probabilities that are calculated according to the Kullback–Leibler divergence between the posterior distribution using the observations previously allocated and the posterior distribution including a ‘new’ observation. We verified the performance of the proposed algorithm on the simulated data and then we illustrate its use on three publicly available real data sets.

Postselection Inference in Structural Equation Modeling

Most statistical inference methods were established under the assumption that the fitted model is known in advance. In practice, however, researchers often obtain their final model by some data-driven selection process. The selection process makes the finally fitted model random, and it also influences the sampling distribution of the estimator. Therefore, implementing naive inference methods may result in wrong conclusions—which is probably a prime source of the reproducibility crisis in psychological science. The present study accommodates three valid state-of-the-art postselection inference methods for structural equation modeling (SEM) from the statistical literature: data splitting (DS), postselection inference (PoSI), and the polyhedral (PH) method. A simulation is conducted to compare the three methods with the commonly used naive procedure under selection events made by L1-penalized SEM. The results show that the naive method often yields incorrect inference, and that the valid methods control the coverage rate in most cases with their own pros and cons. Real world data examples show the practical use of the valid inference methods.

BLP-2LASSO for aggregate discrete choice models with rich covariates

Summary We introduce the BLP-2LASSO model, which augments the classic BLP (Berry, Levinsohn, and Pakes, 1995) random-coefficients logit model to allow for data-driven selection among a high-dimensional set of control variables using the 'double-LASSO' procedure proposed by Belloni, Chernozhukov, and Hansen (2013). Economists often study consumers’ aggregate behaviour across markets choosing from a menu of differentiated products. In this analysis, local demographic characteristics can serve as controls for market-specific preference heterogeneity. Given rich demographic data, implementing these models requires specifying which variables to include in the analysis, an ad hoc process typically guided primarily by a researcher’s intuition. We propose a data-driven approach to estimate these models, applying penalized estimation algorithms from the recent literature in high-dimensional econometrics. Our application explores the effect of campaign spending on vote shares in data from Mexican elections.

Abstract 4014: Clinical validation of Lantern Pharma’s Response Algorithm for Drug Positioning and Rescue (RADRTM)

Abstract Lantern Pharma has developed a technology platform termed RADRTM that can be used to predict true responders before conducting a clinical trial in order to achieve higher success rates. RADRTM is an Artificial Intelligence (Al)-based machine learning approach for complex biomarker identification and patient stratification. RADRTM is a combination of three automated modules working sequentially to generate drug- and tumor-specific gene signatures predictive of response. RADRTM integrates biological knowledge, data-driven feature selection, and robust Al algorithms to facilitate hypothesis-free drug- and cancer-specific biomarker development. We present retrospective analyses performed as part of RADRTM validation using at least 9 independent datasets of patients from selected cancer types treated with approved drugs including chemotherapy, targeted therapy and immune-oncology agents. Pre-treatment patient gene expression profiles along with corresponding treatment outcomes were used as algorithm inputs. Model training was typically performed using an initial set of genes derived from cancer cell line data when available, and further applied to a subset of patient data for model tuning and final gene signature development. Model testing and performance computation were carried out on patient records held out as blinded datasets. The response prediction accuracy, true positive rate (TPR), true negative rate (TNR) false discovery rate, positive predictive value and Matthew’s Correlation Coefficient were among the model performance metrics calculated. On average, RADRTM achieved a response prediction accuracy of 80% during clinical validation. For instance, in an analysis of 92 breast cancer patients, RADRTM generated a signature of 18 genes whose expression level was predictive of Paclitaxel treatment response at an overall accuracy of 78% and 81% TPR/ 76% TNR. The above results imply that the application of the RADRTM program to this Paclitaxel trial in breast cancer patients could have potentially reduced the number of patients in the treatment arm from 92 unselected patients to 24 biomarker-selected patients to produce the same number of responders. Moreover, we cite published evidence correlating genes from this 18-gene signature with increased Paclitaxel sensitivity in breast cancer. The value of the platform architecture is derived from its validation through the analysis of about 6 million oncology-specific clinical data points, more than 120 drug-cancer interactions, and over 600 patient records. Thus, by implementing unique biological, statistical and machine learning workflows, Lantern Pharma's RADRTM technology is capable of deriving robust biomarker panels for pre-selecting true responders for recruitment into clinical trials which may improve the success rate of oncology drug approvals. Citation Format: Yuvanesh Vedaraju, Umesh Kathad, Aditya Kulkarni, Barry Henderson, Gregory Tobin, Panna Sharma, Arun Asaithambi. Clinical validation of Lantern Pharma’s Response Algorithm for Drug Positioning and Rescue (RADRTM) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 4014.

Abstract 672: Establishment of a drug-tumor interaction database using Lantern Pharma’s Response Algorithm for Drug Positioning and Rescue (RADRTM)

Abstract The Response Algorithm for Drug positioning and Rescue (RADRTM) technology is Lantern Pharma's proprietary Artificial Intelligence (Al)-based machine learning approach for biomarker panel identification and patient stratification. RADRTM is a combination of three automated modules working sequentially to generate drug- and tumor-specific gene signatures predictive of response. RADRTM integrates biological knowledge, data-driven feature selection, and robust Al algorithms to facilitate hypothesis-free, drug- and cancer-specific biomarker development. RADRTM uses transcriptomic, drug sensitivity datasets and systems biology inputs and generates gene expression-based responder/non-responder profiles for specific tumor indications with high accuracy. RADRTM uses a unique process flow and a combination of machine learning algorithms to extract drug-specific biomarkers from whole transcriptome level input (~18000 genes). RADRTM comprises three main modules: data pre-processing, feature selection, and response prediction. Data pre-processing includes data cleaning, transformation and normalization. For dimensionality reduction and feature selection, RADRTM performs gene filtering via biological and statistical feature selection methods. In the final component, an AI-driven program reduces the intermediate number (approximately 500) of genes to a more manageable number (10 - 50) of candidate biomarkers capable of predicting drug sensitivity or insensitivity. Lantern’s RADRTM AI application incorporates automated supervised machine learning strategies along with big data analytics, statistics and systems biology to enable identification of new correlations of genetic biomarkers with drug activity. Using RADRTM we have created a database of drug response prediction models for more than 120 drug-tumor type combinations in a preclinical setting that is expected to keep growing. These drug- and cancer-specific RADRTMmodels have further enabled the classification of clinical records into distinct response groups, as well as generated gene expression signatures as features predictive of response. The average response prediction accuracy lies above 80%. We also demonstrate the utility of such a database in clinical translation through various performance metrics including but not limited to true positive rate (TPR), true negative rate (TNR), positive predictive value (PPV), negative predictive value (NPV) and area under the curve (AUC). This database links the majority of FDA approved and selected investigational drugs with appropriate cancer indications and the associated RADRTM-derived responder/non-responder profiles in terms of gene expression signatures. This database could directly inform the drug-companion diagnostic co-developmental pathways for new drugs and cancer indications. Citation Format: Umesh Kathad, Yuvanesh Vedaraju, Aditya Kulkarni, Barry Henderson, Gregory Tobin, Panna Sharma, Arun Asaithambi. Establishment of a drug-tumor interaction database using Lantern Pharma’s Response Algorithm for Drug Positioning and Rescue (RADRTM) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 672.

Abstract 4789: Predicting sensitivity to Lantern Pharma’s pipeline drug candidate LP-184 using the Response Algorithm for Drug Positioning and Rescue

Abstract LP-184 is a DNA Damage Repair inhibitor being developed by Lantern Pharma primarily as a non-hormone, non-chemotherapy option for the growing indication of taxane- and hormone-resistant metastatic prostate cancer. LP-184 is a next-generation analog of Irofulven with broad anti-tumor activity that counteracts multi-drug resistance pathways and is potentially synergistic with many classes of anticancer agents. LP-184 has a favorable therapeutic index and pharmacokinetic profile. Knowledge about its shared mechanism of action with Irofulven and potential biomarkers implicated in induction of bioactivation and synthetic lethal interactions is available. To advance LP-184 into clinical stages and achieve accelerated approval, Lantern Pharma is employing its proprietary Artificial Intelligence (AI)-driven technology. Lantern Pharma has developed a technology platform termed RADRTM that can be used to construct responder/ non-responder profiles based on gene expression signatures and predict true responders before conducting a clinical trial in order to achieve higher success rates. RADRTM is an Al-based machine learning approach for candidate biomarker identification and patient stratification. RADRTM is a combination of three automated modules working sequentially to generate drug- and tumor-specific gene signatures predictive of response. RADRTM emphasizes the integration of biological knowledge, data-driven feature selection, and robust Al algorithms to derive biomarkers in a hypothesis-free manner. In analytic demonstrations, RADRTM was able to achieve an average accuracy of 85% in validation tests using preclinical datasets associated with selected solid tumor indications and approved drugs. As part of RADRTM drug model building and development, we used a dataset showing preclinical efficacy of our pipeline oncology candidate LP-184. We obtained baseline cell line gene expression profiles covering more than 18,000 transcripts per cell line and corresponding LP-184 sensitivity records from the NCI60 cancer cell line panel. Using RADRTM, we derived a panel of 10 genes whose expression levels are predictive of overall response at an accuracy of 100%. Thus, RADRTM was able to identify the top 10 genes for prediction of either drug sensitivity or insensitivity, demonstrating the hypothesis-free identification of biomarkers with biological relevance and statistical rigor and having highest possible prediction accuracy. Genes from the final 10 predictive list were found to be functionally involved in LP-184-specific induction of bioactivation and are in agreement with its mechanism of action. These preliminary biomarker analyses will be further validated using LP-184 sensitivity and pre-treatment gene expression data derived from ex vivo models of fresh prostate tumor biopsy samples. Citation Format: Aditya Kulkarni, Umesh Kathad, Yuvanesh Vedaraju, Barry Henderson, Gregory Tobin, Panna Sharma, Arun Asaithambi. Predicting sensitivity to Lantern Pharma’s pipeline drug candidate LP-184 using the Response Algorithm for Drug Positioning and Rescue [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 4789.

Exploring intrinsic triggers for functional facial electrostimulation based on intramuscular electromyography recordings.

Based on univariate intramuscular electromyography (EMG) recordings of facial muscles of patients suffering from chronic idiopathic facial palsy we propose a data-driven feature selection process for the discrimination of different mimic maneuvers. Following fundamental ideas of automatic EMG decompositions based on templates defined by motor unit action potentials, the proposed approach relies on a multiple template matching. Yet, the novel methodology utilizes templates derived from the intramuscular EMG signal itself without any supervisor interaction or a priori information by identifying abundant short signal sections (motifs). Focusing on motifs as individual, characteristical graphoelements of an EMG recording implies a high level of flexibility. In connection with facial palsy such a flexibility is necessary, since unique individual, also pathological, EMG patterns can be expected due to the high spatial variability of intramuscular recordings combined with random patterns of aberrant reinnervation. The proposed methodology is applied to EMG data of frontalis, zygomaticus, and orbicularis oculi muscle without patient- or muscle-specific adaptations.

Penalized I-spline monotone regression estimation

We propose a penalized regression spline estimator for monotone regression. To construct the estimator, we adopt the I-splines with the total variation penalty. The I-splines lend themselves to the monotonicity because of the simpler form of restrictions, and the total variation penalty induces a data-driven knot selection scheme. A coordinate descent algorithm is developed for the estimator. If the number of complexity parameter candidates sufficiently increases, the algorithm considers all possible monotone linear spline fits to the given data. The pruning process of the algorithm not only provides numerical stability, but also implements the data-driven knot selection. We also compute the maximum candidate of the complexity parameter to facilitate complexity parameter selection. Extensive numerical studies show that the proposed estimator captures spatially inhomogeneous behaviors of data, such as sudden jumps.

O-43 Data-driven selection of active iEEG channels during verbal memory task performance

Background Specific brain regions activated during a memory task can be accurately determined using spectral power and amplitude from intracranial electroencephalography (iEEG) signals sampled from the overlying electrodes implanted across the neocortex. We developed a data-driven method for active electrode determination and compared our results with a previously used approach based on a predefined arbitrary threshold. Material and methods We assessed neurophysiological activity of 57 epilepsy patients during a verbal short-term memory task, in which lists of 12 common nouns were presented for subsequent freely recall. Normalized amplitude of the high gamma frequency component of the signal was extracted from the word presentation epochs. A family of exponential functions was used to automatically determine patient-specific threshold values for selecting active channels. Results Proportion of active electrodes determined with our data-driven method was lower (2056 of 5857 channels – 35.28%) than for the method using a predefined threshold (3865 channels – 66.33%). There was 67.70% overlap between the channels selected with the two. Methods We observed significant differences between the two populations of selected active electrodes obtained with these methods, including the profile of average spectral power response during word presentation and memory effect differences between words that were subsequently recalled and forgotten. Conclusions The proposed method offers a fully automatic, patient-specific selection of active electrodes sampling from brain areas activated by verbal memory processing. Our results suggest improved sensitivity and specificity of active electrode selection that can potentially be applied to other cognitive, perceptual, and motor tasks.

An evaluation of likelihood-based bandwidth selectors for spatial and spatiotemporal kernel estimates

ABSTRACTSpatial point pattern data sets are commonplace in a variety of different research disciplines. The use of kernel methods to smooth such data is a flexible way to explore spatial trends and make inference about underlying processes without, or perhaps prior to, the design and fitting of more intricate semiparametric or parametric models to quantify specific effects. The long-standing issue of ‘optimal’ data-driven bandwidth selection is complicated in these settings by issues such as high heterogeneity in observed patterns and the need to consider edge correction factors. We scrutinize bandwidth selectors built on leave-one-out cross-validation approximation to likelihood functions. A key outcome relates to previously unconsidered adaptive smoothing regimens for spatiotemporal density and multitype conditional probability surface estimation, whereby we propose a novel simultaneous pilot-global selection strategy. Motivated by applications in epidemiology, the results of both simulated and real-world analyses suggest this strategy to be largely preferable to classical fixed-bandwidth estimation for such data.

Questioning the news about economic growth: Sparse forecasting using thousands of news-based sentiment values

The modern calculation of textual sentiment involves a myriad of choices as to the actual calibration. We introduce a general sentiment engineering framework that optimizes the design for forecasting purposes. It includes the use of the elastic net for sparse data-driven selection and the weighting of thousands of sentiment values. These values are obtained by pooling the textual sentiment values across publication venues, article topics, sentiment construction methods, and time. We apply the framework to the investigation of the value added by textual analysis-based sentiment indices for forecasting economic growth in the US. We find that the additional use of optimized news-based sentiment values yields significant accuracy gains for forecasting the nine-month and annual growth rates of the US industrial production, compared to the use of high-dimensional forecasting techniques based on only economic and financial indicators.

Data-adaptive trimming of the Hill estimator and detection of outliers in the extremes of heavy-tailed data

We introduce a trimmed version of the Hill estimator for the index of a heavy-tailed distribution, which is robust to perturbations in the extreme order statistics. In the ideal Pareto setting, the estimator is essentially finite-sample efficient among all unbiased estimators with a given strict upper break-down point. For general heavy-tailed models, we establish the asymptotic normality of the estimator under second order regular variation conditions and also show that it is minimax rate-optimal in the Hall class of distributions. We also develop an automatic, data-driven method for the choice of the trimming parameter which yields a new type of robust estimator that can adapt to the unknown level of contamination in the extremes. This adaptive robustness property makes our estimator particularly appealing and superior to other robust estimators in the setting where the extremes of the data are contaminated. As an important application of the data-driven selection of the trimming parameters, we obtain a methodology for the principled identification of extreme outliers in heavy tailed data. Indeed, the method has been shown to correctly identify the number of outliers in the previously explored Condroz data set.

A new approach to estimator selection

In the framework of an abstract statistical model, we discuss how to use the solution of one estimation problem (Problem A) in order to construct an estimator in another, completely different, Problem B. As a solution of Problem A we understand a data-driven selection from a given family of estimators $\mathbf{A}(\mathfrak{H})=\{\widehat{A}_{\mathfrak{h}},\mathfrak{h}\in\mathfrak{H}\}$ and establishing for the selected estimator so-called oracle inequality. If $\hat{\mathfrak{h}}\in\mathfrak{H}$ is the selected parameter and $\mathbf{B}(\mathfrak{H})=\{\widehat{B}_{\mathfrak{h}},\mathfrak{h}\in\mathfrak{H}\}$ is an estimator’s collection built in Problem B, we suggest to use the estimator $\widehat{B}_{\hat{\mathfrak{h}}}$. We present very general selection rule led to selector $\hat{\mathfrak{h}}$ and find conditions under which the estimator $\widehat{B}_{\hat{\mathfrak{h}}}$ is reasonable. Our approach is illustrated by several examples related to adaptive estimation.

Data-Driven Lightweight Interest Point Selection for Large-Scale Visual Search

With the explosive increase of images and videos, visual analysis has become an essential technique in dealing with the big visual data, which utilizes the visual feature descriptors to search or recognize the images or frames with target objects or events. Subject to the constraints of resources (e.g., memory, bandwidth, storage, etc.), interest point selection is crucial to generate robust compact descriptors for high-efficiency visual analysis by selecting and aggregating the most discriminative local feature descriptors, which has been demonstrated in the state-of-the-art low bit rate visual search works. In this paper, we propose a data-driven lightweight interest point selection approach to significantly improve the performance of visual search, while ameliorating the efficiency of extracting feature descriptors. Comprehensive experimental results over benchmarks have shown that the proposed interest point selection algorithm has significantly improved image matching and retrieval performance in the completed MPEG Compact Descriptors for Visual Search (CDVS) standard as well as the emerging MPEG Compact Descriptors for Video Analytics (CDVA) standard, say 20% mAP gain by data-driven selection against random selection of interest points. In particular, the presented data-driven interest point selection has been adopted by MPEG-CDVS and MPEG-CDVA as a normative technique to improve the aggregation of handcrafted features, which has contributed to the combination of handcrafted features and deep learning (CNN) features as well.

Data-Driven Bandwidth Selection for Recursive Kernel Density Estimators Under Double Truncation

In this paper we proposed a data-driven bandwidth selection procedure of the recursive kernel density estimators under double truncation. We showed that, using the selected bandwidth and a special stepsize, the proposed recursive estimators outperform the nonrecursive one in terms of estimation error in many situations. We corroborated these theoretical results through simulation study. The proposed estimators are then applied to data on the luminosity of quasars in astronomy. We corroborated these theoretical results through simulation study, then, we applied the proposed estimators to data on the luminosity of quasars in astronomy.

New robust statistical procedures for the polytomous logistic regression models.

This article derives a new family of estimators, namely the minimum density power divergence estimators, as a robust generalization of the maximum likelihood estimator for the polytomous logistic regression model. Based on these estimators, a family of Wald-type test statistics for linear hypotheses is introduced. Robustness properties of both the proposed estimators and the test statistics are theoretically studied through the classical influence function analysis. Appropriate real life examples are presented to justify the requirement of suitable robust statistical procedures in place of the likelihood based inference for the polytomous logistic regression model. The validity of the theoretical results established in the article are further confirmed empirically through suitable simulation studies. Finally, an approach for the data-driven selection of the robustness tuning parameter is proposed with empirical justifications.