Multi-step adaptive elastic-net: reducing false positives in high-dimensional variable selection

Resting-State Functional Connectivity Predicts Attention Problems in Children: Evidence from the ABCD Study

Attention deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder, and numerous functional and structural differences have been identified in the brains of individuals with ADHD compared to controls. This study uses data from the baseline sample of the large, epidemiologically informed Adolescent Brain Cognitive Development Study of children aged 9-10 years old (N = 7979). Cross-validated Poisson elastic net regression models were used to predict a dimensional measure of ADHD symptomatology from within- and between-network resting-state correlations and several known risk factors, such as biological sex, socioeconomic status, and parental history of problematic alcohol and drug use. We found parental history of drug use and biological sex to be the most important predictors of attention problems. The connection between the default mode network and the dorsal attention network was the only brain network identified as important for predicting attention problems. Specifically, we found that reduced magnitudes of the anticorrelation between the default mode and dorsal attention networks relate to increased attention problems in children. Our findings complement and extend recent studies that have connected individual differences in structural and task-based fMRI to ADHD symptomatology and individual differences in resting-state fMRI to ADHD diagnoses.

Robust variable selection methods with Cox model-a selective practical benchmark study

With the advancement of biological and medical techniques, we can now obtain large amounts of high-dimensional omics data with censored survival information. This presents challenges in method development across various domains, particularly in variable selection. Given the inherently skewed distribution of the survival time outcome variable, robust variable selection methods offer potential solutions. Recently, there has been a focus on extending robust variable selection methods from linear regression models to survival models. However, despite these developments, robust methods are currently rarely used in practical applications, possibly due to a limited appreciation of their overall good performance. To address this gap, we conduct a selective review comparing the variable selection performance of twelve robust and non-robust penalised Cox models. Our study reveals the intricate relationship among covariates, survival outcomes, and modeling approaches, demonstrating how subtle variations can significantly impact the performance of methods considered. Based on our empirical research, we recommend the use of robust Cox models for variable selection in practice based on their superior performance in presence of outliers while maintaining good efficiency and accuracy when there are no outliers. This study provides valuable insights for method development and application, contributing to a better understanding of the relationship between correlated covariates and censored outcomes.

Omics feature selection with the extended SIS R package: identification of a body mass index epigenetic multimarker in the Strong Heart Study

The statistical analysis of omics data poses a great computational challenge given their ultra-high-dimensional nature and frequent between-features correlation. In this work, we extended the iterative sure independence screening (ISIS) algorithm by pairing ISIS with elastic-net (Enet) and 2 versions of adaptive elastic-net (adaptive elastic-net (AEnet) and multistep adaptive elastic-net (MSAEnet)) to efficiently improve feature selection and effect estimation in omics research. We subsequently used genome-wide human blood DNA methylation data from American Indian participants in the Strong Heart Study (n = 2235 participants; measured in 1989-1991) to compare the performance (predictive accuracy, coefficient estimation, and computational efficiency) of ISIS-paired regularization methods with that of a bayesian shrinkage and traditional linear regression to identify an epigenomic multimarker of body mass index (BMI). ISIS-AEnet outperformed the other methods in prediction. In biological pathway enrichment analysis of genes annotated to BMI-related differentially methylated positions, ISIS-AEnet captured most of the enriched pathways in common for at least 2 of all the evaluated methods. ISIS-AEnet can favor biological discovery because it identifies the most robust biological pathways while achieving an optimal balance between bias and efficient feature selection. In the extended SIS R package, we also implemented ISIS paired with Cox and logistic regression for time-to-event and binary endpoints, respectively, and a bootstrap approach for the estimation of regression coefficients.

The impact of Bayesian optimization on feature selection

Feature selection is an indispensable step for the analysis of high-dimensional molecular data. Despite its importance, consensus is lacking on how to choose the most appropriate feature selection methods, especially when the performance of the feature selection methods itself depends on hyper-parameters. Bayesian optimization has demonstrated its advantages in automatically configuring the settings of hyper-parameters for various models. However, it remains unclear whether Bayesian optimization can benefit feature selection methods. In this research, we conducted extensive simulation studies to compare the performance of various feature selection methods, with a particular focus on the impact of Bayesian optimization on those where hyper-parameters tuning is needed. We further utilized the gene expression data obtained from the Alzheimer's Disease Neuroimaging Initiative to predict various brain imaging-related phenotypes, where various feature selection methods were employed to mine the data. We found through simulation studies that feature selection methods with hyper-parameters tuned using Bayesian optimization often yield better recall rates, and the analysis of transcriptomic data further revealed that Bayesian optimization-guided feature selection can improve the accuracy of disease risk prediction models. In conclusion, Bayesian optimization can facilitate feature selection methods when hyper-parameter tuning is needed and has the potential to substantially benefit downstream tasks.

Genomic prediction using machine learning: a comparison of the performance of regularized regression, ensemble, instance-based and deep learning methods on synthetic and empirical data

BACKGROUND

The accurate prediction of genomic breeding values is central to genomic selection in both plant and animal breeding studies. Genomic prediction involves the use of thousands of molecular markers spanning the entire genome and therefore requires methods able to efficiently handle high dimensional data. Not surprisingly, machine learning methods are becoming widely advocated for and used in genomic prediction studies. These methods encompass different groups of supervised and unsupervised learning methods. Although several studies have compared the predictive performances of individual methods, studies comparing the predictive performance of different groups of methods are rare. However, such studies are crucial for identifying (i) groups of methods with superior genomic predictive performance and assessing (ii) the merits and demerits of such groups of methods relative to each other and to the established classical methods. Here, we comparatively evaluate the genomic predictive performance and informally assess the computational cost of several groups of supervised machine learning methods, specifically, regularized regression methods, deep, ensemble and instance-based learning algorithms, using one simulated animal breeding dataset and three empirical maize breeding datasets obtained from a commercial breeding program.

RESULTS

Our results show that the relative predictive performance and computational expense of the groups of machine learning methods depend upon both the data and target traits and that for classical regularized methods, increasing model complexity can incur huge computational costs but does not necessarily always improve predictive accuracy. Thus, despite their greater complexity and computational burden, neither the adaptive nor the group regularized methods clearly improved upon the results of their simple regularized counterparts. This rules out selection of one procedure among machine learning methods for routine use in genomic prediction. The results also show that, because of their competitive predictive performance, computational efficiency, simplicity and therefore relatively few tuning parameters, the classical linear mixed model and regularized regression methods are likely to remain strong contenders for genomic prediction.

CONCLUSIONS

The dependence of predictive performance and computational burden on target datasets and traits call for increasing investments in enhancing the computational efficiency of machine learning algorithms and computing resources.

Exploring the gut microbiota in patients with pre-diabetes and treatment naïve diabetes type 2 - a pilot study

BACKGROUND

Compared to their healthy counterparts, patients with type 2 diabetes (T2D) can exhibit an altered gut microbiota composition, correlated with detrimental outcomes, including reduced insulin sensitivity, dyslipidemia, and increased markers of inflammation. However, a typical T2D microbiota profile is not established. The aim of this pilot study was to explore the gut microbiota and bacteria associated with prediabetes (pre-T2D) patients, and treatment naïve T2D patients, compared to healthy subjects.

METHODS

Fecal samples were collected from patients and healthy subjects (from Norway). The bacterial genomic DNA was extracted, and the microbiota analyzed utilizing the bacterial 16S rRNA gene. To secure a broad coverage of potential T2D associated bacteria, two technologies were used: The GA-map® 131-plex, utilizing 131 DNA probes complementary to pre-selected bacterial targets (covering the 16S regions V3-V9), and the LUMI-Seq™ platform, a full-length 16S sequencing technology (V1-V9). Variations in the gut microbiota between groups were explored using multivariate methods, differential bacterial abundance was estimated, and microbiota signatures discriminating the groups were assessed using classification models.

RESULTS

In total, 24 pre-T2D patients, 18 T2D patients, and 52 healthy subjects were recruited. From the LUMI-Seq™ analysis, 10 and 9 bacterial taxa were differentially abundant between pre-T2D and healthy, and T2D and healthy, respectively. From the GA-map® 131-plex analysis, 10 bacterial markers were differentially abundant when comparing pre-T2D and healthy. Several of the bacteria were short-chain fatty acid (SCFA) producers or typical opportunistic bacteria. Bacteria with similar function or associated properties also contributed to the separation of pre-T2D and T2D from healthy as found by classification models. However, limited overlap was found for specific bacterial genera and species.

CONCLUSIONS

This pilot study revealed that differences in the abundance of SCFA producing bacteria, and an increase in typical opportunistic bacteria, may contribute to the variations in the microbiota separating the pre-T2D and T2D patients from healthy subjects. However, further efforts in investigating the relationship between gut microbiota, diabetes, and associated factors such as BMI, are needed for developing specific diabetes microbiota signatures.

New index of functional specificity to predict the redundancy of ecosystem functions in microbial communities

An ecosystem function is suggested to be more sensitive to biodiversity loss (i.e. low functional redundancy) when focusing on specific-type functions than broad-type functions. Thus far, specific-type functions have been loosely defined as functions performed by a small number of species (facilitative species) or functions involved in utilizing complex substrates. However, quantitative examination of functional specificity remains underexplored. We quantified the functional redundancy of 33 ecosystem functions in a freshwater system from 76 prokaryotic community samples over three years. For each function, we used a sparse regression model to estimate the number of facilitative Amplicon Sequence Variants (ASVs) and to define taxon-based functional specificity. We also used Bertz structural complexity to determine substrate-based functional specificity. We found that functional redundancy increased with the taxon-based functional specificity defined as the proportion of facilitative ASVs ( = facilitative ASV richness/ facilitative ASV richness + repressive ASV (ASVs reducing functioning) richness). When using substrate-based functional specificity, functional redundancy was influenced by Bertz complexity per se and by substrate acquisition mechanisms. Therefore, taxon-based functional specificity is a better predictive index for evaluating functional redundancy than substrate-based functional specificity. These findings provide a framework to quantitatively predict the consequences of diversity losses on ecosystem functioning.

Predictive nomograms for oral tongue squamous cell carcinoma applying the American Joint Committee on Cancer/Union Internationale Contre le Cancer 8th edition staging system

BACKGROUND

Nomograms applying the 8th edition of the TNM staging system aimed at predicting overall (OS), disease-specific (DSS), locoregional recurrence-free (LRRFS) and distant recurrence-free survivals (DRFS) for oral tongue squamous cell carcinoma (OTSCC) are still lacking.

METHODS

A training cohort of 438 patients with OTSCC was retrospectively enrolled from a single institution. An external validation set of 287 patients was retrieved from two independent institutions.

RESULTS

Internal validation of the multivariable models for OS, DSS, DRFS and LRRFS showed a good calibration and discrimination results with optimism-corrected c-indices of 0.74, 0.75, 0.77 and 0.70, respectively. The external validation confirmed the good performance of OS, DSS and DRFS models (c-index 0.73 and 0.77, and 0.73, respectively) and a fair performance of the LRRFS model (c-index 0.58).

CONCLUSIONS

The nomograms herein presented can be implemented as useful tools for prediction of OS, DSS, DRFS and LRRFS in OTSCC.

The Error-Related Negativity Predicts Self-Control Failures in Daily Life

Adaptive behavior critically depends on performance monitoring (PM), the ability to monitor action outcomes and the need to adapt behavior. PM-related brain activity has been linked to guiding decisions about whether action adaptation is warranted. The present study examined whether PM-related brain activity in a flanker task, as measured by electroencephalography (EEG), was associated with adaptive behavior in daily life. Specifically, we were interested in the employment of self-control, operationalized as self-control failures (SCFs), and measured using ecological momentary assessment. Analyses were conducted using an adaptive elastic net regression to predict SCFs from EEG in a sample of 131 participants. The model was fit using within-subject averaged response-locked EEG activity at each electrode and time point within an epoch surrounding the response. We found that higher amplitudes of the error-related negativity (ERN) were related to fewer SCFs. This suggests that lower error-related activity may relate to lower recruitment of interventive self-control in daily life. Altered cognitive control processes, like PM, have been proposed as underlying mechanisms for various mental disorders. Understanding how alterations in PM relate to regulatory control might therefore aid in delineating how these alterations contribute to different psychopathologies.

Structure and process associated with the efficiency of intensive care units in low-resource settings: An analysis of the CHECKLIST-ICU trial database

PURPOSE

Characteristics of structure and process impact ICU performance and the outcomes of critically ill patients. We sought to identify organizational characteristics associated with efficient ICUs in low-resource settings.

MATERIALS AND METHODS

This is a secondary analysis of a multicenter cluster-randomized clinical trial in Brazil (CHECKLIST-ICU). Efficient units were defined by standardized mortality ratio (SMR) and standardized resource use (SRU) lower than the overall medians and non-efficient otherwise. We used a regularized logistic regression model to evaluate associations between organizational factors and efficiency.

RESULTS

From 118 ICUs (13,635 patients), 47 units were considered efficient and 71 non-efficient. Efficient units presented lower incidence rates (median[IQR]) of central line-associated bloodstream infections (4.95[0.00-22.0] vs 6.29[0.00-25.6], p = .04), utilization rates of mechanical ventilation (0.41[0.07-0.73] vs 0.58[0.19-0.82], p < .001), central venous catheter (0.67[0.15-0.98] vs 0.78[0.33-0.98], p = .04), and indwelling urinary catheter (0.62[0.22-0.95] vs 0.76[0.32-0.98], p < .01) than non-efficient units. The reported active surveillance of ventilator-associated pneumonia (OR = 1.72; 95%CI, 1.16-2.57) and utilization of central venous catheters (OR = 1.94; 95%CI, 1.32-2.94) were associated with efficient ICUs.

CONCLUSIONS

In low-resource settings, active surveillance of nosocomial infections and the utilization of invasive devices were associated with efficiency, supporting the management and evaluation of performance indicators as a starting point for improvement in ICU.

Variable selection for inferential models with relatively high-dimensional data: Between method heterogeneity and covariate stability as adjuncts to robust selection

Variable selection in inferential modelling is problematic when the number of variables is large relative to the number of data points, especially when multicollinearity is present. A variety of techniques have been described to identify 'important' subsets of variables from within a large parameter space but these may produce different results which creates difficulties with inference and reproducibility. Our aim was evaluate the extent to which variable selection would change depending on statistical approach and whether triangulation across methods could enhance data interpretation. A real dataset containing 408 subjects, 337 explanatory variables and a normally distributed outcome was used. We show that with model hyperparameters optimised to minimise cross validation error, ten methods of automated variable selection produced markedly different results; different variables were selected and model sparsity varied greatly. Comparison between multiple methods provided valuable additional insights. Two variables that were consistently selected and stable across all methods accounted for the majority of the explainable variability; these were the most plausible important candidate variables. Further variables of importance were identified from evaluating selection stability across all methods. In conclusion, triangulation of results across methods, including use of covariate stability, can greatly enhance data interpretation and confidence in variable selection.

Rapid Identification of Rainbow Trout Adulteration in Atlantic Salmon by Raman Spectroscopy Combined with Machine Learning

This study intends to evaluate the utilization potential of the combined Raman spectroscopy and machine learning approach to quickly identify the rainbow trout adulteration in Atlantic salmon. The adulterated samples contained various concentrations (0-100% w/w at 10% intervals) of rainbow trout mixed into Atlantic salmon. Spectral preprocessing methods, such as first derivative, second derivative, multiple scattering correction (MSC), and standard normal variate, were employed. Unsupervised algorithms, such as recursive feature elimination, genetic algorithm (GA), and simulated annealing, and supervised K-means clustering (KM) algorithm were used for selecting important spectral bands to reduce the spectral complexity and improve the model stability. Finally, the performances of various machine learning models, including linear regression, nonlinear regression, regression tree, and rule-based models, were verified and compared. The results denoted that the developed GA-KM-Cubist machine learning model achieved satisfactory results based on MSC preprocessing. The determination coefficient (R2) and root mean square error of prediction sets (RMSEP) in the test sets were 0.87 and 10.93, respectively. These results indicate that Raman spectroscopy can be used as an effective Atlantic salmon adulteration identification method; further, the developed model can be used for quantitatively analyzing the rainbow trout adulteration in Atlantic salmon.

Predicting patient-reported outcomes following hip and knee replacement surgery using supervised machine learning

BACKGROUND

Machine-learning classifiers mostly offer good predictive performance and are increasingly used to support shared decision-making in clinical practice. Focusing on performance and practicability, this study evaluates prediction of patient-reported outcomes (PROs) by eight supervised classifiers including a linear model, following hip and knee replacement surgery.

METHODS

NHS PRO data (130,945 observations) from April 2015 to April 2017 were used to train and test eight classifiers to predict binary postoperative improvement based on minimal important differences. Area under the receiver operating characteristic, J-statistic and several other metrics were calculated. The dependent outcomes were generic and disease-specific improvement based on the EQ-5D-3L visual analogue scale (VAS) as well as the Oxford Hip and Knee Score (Q score).

RESULTS

The area under the receiver operating characteristic of the best training models was around 0.87 (VAS) and 0.78 (Q score) for hip replacement, while it was around 0.86 (VAS) and 0.70 (Q score) for knee replacement surgery. Extreme gradient boosting, random forests, multistep elastic net and linear model provided the highest overall J-statistics. Based on variable importance, the most important predictors for post-operative outcomes were preoperative VAS, Q score and single Q score dimensions. Sensitivity analysis for hip replacement VAS evaluated the influence of minimal important difference, patient selection criteria as well as additional data years. Together with a small benchmark of the NHS prediction model, robustness of our results was confirmed.

CONCLUSIONS

Supervised machine-learning implementations, like extreme gradient boosting, can provide better performance than linear models and should be considered, when high predictive performance is needed. Preoperative VAS, Q score and specific dimensions like limping are the most important predictors for postoperative hip and knee PROMs.