STatistical Inference Relief (STIR) feature selection

An integrated deep learning approach for modeling dissolved oxygen concentration at coastal inlets based on hydro-climatic parameters

Climate change has a significant impact on dissolved oxygen (DO) concentrations, particularly in coastal inlets where numerous human activities occur. Due to the various water quality (WQ), hydrological, and climatic parameters that influence this phenomenon, predicting and modeling DO variation is a challenging process. Accordingly, this study introduces an innovative Deep Learning Neural Network (DLNN) methodology to model and predict DO concentrations for the Egyptian Rashid coastal inlet, leveraging field-recorded WQ and hydroclimatic datasets. Initially, statistical and exploratory data analyses are performed to provide a thorough understanding of the relationship between DO fluctuations and associated WQ and hydroclimatic stressors. As an initial step towards developing an effective DO predictive model, conventional Machine Learning (ML) approaches such as Gaussian Process Regression (GPR), Support Vector Regression (SVR), and Decision Tree Regressor (DTR) are employed. Subsequently, a DLNN approach is utilized to validate the prediction capabilities of the investigated conventional ML approaches. Finally, a sensitivity analysis is conducted to evaluate the impact of WQ and hydroclimatic parameters on predicted DO. The outcomes demonstrate that DLNN significantly improves DO prediction accuracy by 4% compared to the best-performing ML approach, achieving a Correlation Coefficient of 0.95 with a root mean square error (RMSE) of 0.42 mg/l. Solar radiation (SR), pH, water levels (WL), and atmospheric pressure (P) emerge as the most significant hydroclimatic parameters influencing DO fluctuations. Ultimately, the developed models could serve as effective indicators for coastal authorities to monitor DO changes resulting from accelerated climate change along the Egyptian coast.

Individualized treatment decision model for inoperable elderly esophageal squamous cell carcinoma based on multi-modal data fusion

BACKGROUND

This research aimed to develop a model for individualized treatment decision-making in inoperable elderly patients with esophageal squamous cell carcinoma (ESCC) using machine learning methods and multi-modal data.

METHODS

A total of 189 inoperable elderly ESCC patients aged 65 or older who underwent concurrent chemoradiotherapy (CCRT) or radiotherapy (RT) were included. Multi-task learning models were created using machine learning techniques to analyze multi-modal data, including pre-treatment CT images, clinical information, and blood test results. Nomograms were constructed to predict the objective response rate (ORR) and progression-free survival (PFS) for different treatment strategies. Optimal treatment plans were recommended based on the nomograms. Patients were stratified into high-risk and low-risk groups using the nomograms, and survival analysis was performed using Kaplan-Meier curves.

RESULTS

The identified risk factors influencing ORR were histologic grade (HG), T stage and three radiomic features including original shape elongation, first-order skewness and original shape flatness, while risk factors influencing PFS included BMI, HG and three radiomic features including high gray-level run emphasis, first-order minimum and first-order skewness. These risk factors were incorporated into the nomograms as independent predictive factors. PFS was substantially different between the low-risk group (total score ≤ 110) and the high-risk group (total score > 110) according to Kaplan-Meier curves (P < 0.05).

CONCLUSIONS

The developed predictive models for ORR and PFS in inoperable elderly ESCC patients provide valuable insights for predicting treatment efficacy and prognosis. The nomograms enable personalized treatment decision-making and can guide optimal treatment plans for inoperable elderly ESCC patients.

An Efficient Feature Selection Algorithm for Gene Families Using NMF and ReliefF

Gene families, which are parts of a genome's information storage hierarchy, play a significant role in the development and diversity of multicellular organisms. Several studies have focused on the characteristics of gene families, such as function, homology, or phenotype. However, statistical and correlation analyses on the distribution of gene family members in the genome have yet to be conducted. Here, a novel framework incorporating gene family analysis and genome selection based on NMF-ReliefF is reported. Specifically, the proposed method starts by obtaining gene families from the TreeFam database and determining the number of gene families within the feature matrix. Then, NMF-ReliefF is used to select features from the gene feature matrix, which is a new feature selection algorithm that overcomes the inefficiencies of traditional methods. Finally, a support vector machine is utilized to classify the acquired features. The results show that the framework achieved an accuracy of 89.1% and an AUC of 0.919 on the insect genome test set. We also employed four microarray gene data sets to evaluate the performance of the NMF-ReliefF algorithm. The outcomes show that the proposed method may strike a delicate balance between robustness and discrimination. Additionally, the proposed method's categorization is superior to state-of-the-art feature selection approaches.

Bi-dimensional principal gene feature selection from big gene expression data

Gene expression sample data, which usually contains massive expression profiles of genes, is commonly used for disease related gene analysis. The selection of relevant genes from huge amount of genes is always a fundamental process in applications of gene expression data. As more and more genes have been detected, the size of gene expression data becomes larger and larger; this challenges the computing efficiency for extracting the relevant and important genes from gene expression data. In this paper, we provide a novel Bi-dimensional Principal Feature Selection (BPFS) method for efficiently extracting critical genes from big gene expression data. It applies the principal component analysis (PCA) method on sample and gene domains successively, aiming at extracting the relevant gene features and reducing redundancies while losing less information. The experimental results on four real-world cancer gene expression datasets show that the proposed BPFS method greatly reduces the data size and achieves a nearly double processing speed compared to the counterpart methods, while maintaining better accuracy and effectiveness.

Opposition-based sine cosine optimizer utilizing refraction learning and variable neighborhood search for feature selection

This paper proposes new improved binary versions of the Sine Cosine Algorithm (SCA) for the Feature Selection (FS) problem. FS is an essential machine learning and data mining task of choosing a subset of highly discriminating features from noisy, irrelevant, high-dimensional, and redundant features to best represent a dataset. SCA is a recent metaheuristic algorithm established to emulate a model based on sine and cosine trigonometric functions. It was initially proposed to tackle problems in the continuous domain. The SCA has been modified to Binary SCA (BSCA) to deal with the binary domain of the FS problem. To improve the performance of BSCA, three accumulative improved variations are proposed (i.e., IBSCA1, IBSCA2, and IBSCA3) where the last version has the best performance. IBSCA1 employs Opposition Based Learning (OBL) to help ensure a diverse population of candidate solutions. IBSCA2 improves IBSCA1 by adding Variable Neighborhood Search (VNS) and Laplace distribution to support several mutation methods. IBSCA3 improves IBSCA2 by optimizing the best candidate solution using Refraction Learning (RL), a novel OBL approach based on light refraction. For performance evaluation, 19 real-wold datasets, including a COVID-19 dataset, were selected with different numbers of features, classes, and instances. Three performance measurements have been used to test the IBSCA versions: classification accuracy, number of features, and fitness values. Furthermore, the performance of the last variation of IBSCA3 is compared against 28 existing popular algorithms. Interestingly, IBCSA3 outperformed almost all comparative methods in terms of classification accuracy and fitness values. At the same time, it was ranked 15 out of 19 in terms of number of features. The overall simulation and statistical results indicate that IBSCA3 performs better than the other algorithms.

Development of a probability calculator for psychosis risk in children, adolescents, and young adults

BACKGROUND

Assessment of risks of illnesses has been an important part of medicine for decades. We now have hundreds of 'risk calculators' for illnesses, including brain disorders, and these calculators are continually improving as more diverse measures are collected on larger samples.

METHODS

We first replicated an existing psychosis risk calculator and then used our own sample to develop a similar calculator for use in recruiting 'psychosis risk' enriched community samples. We assessed 632 participants age 8-21 (52% female; 48% Black) from a community sample with longitudinal data on neurocognitive, clinical, medical, and environmental variables. We used this information to predict psychosis spectrum (PS) status in the future. We selected variables based on lasso, random forest, and statistical inference relief; and predicted future PS using ridge regression, random forest, and support vector machines.

RESULTS

Cross-validated prediction diagnostics were obtained by building and testing models in randomly selected sub-samples of the data, resulting in a distribution of the diagnostics; we report the mean. The strongest predictors of later PS status were the Children's Global Assessment Scale; delusions of predicting the future or having one's thoughts/actions controlled; and the percent married in one's neighborhood. Random forest followed by ridge regression was most accurate, with a cross-validated area under the curve (AUC) of 0.67. Adjustment of the model including only six variables reached an AUC of 0.70.

CONCLUSIONS

Results support the potential application of risk calculators for screening and identification of at-risk community youth in prospective investigations of developmental trajectories of the PS.

An Approach of Epistasis Detection Using Integer Linear Programming Optimizing Bayesian Network

Proposing a more effective and accurate epistatic loci detection method in large-scale genomic data has important research significance for improving crop quality, disease treatment, etc. Due to the characteristics of high accuracy and processing non-linear relationship, Bayesian network (BN) has been widely used in constructing the network of SNPs and phenotype traits and thus to mine epistatic loci. However, the shortcoming of BN is that it is easy to fall into local optimum and unable to process large-scale of SNPs. In this work, we transform the problem of learning Bayesian network into the optimization of integer linear programming (ILP). We use the algorithms of branch-and-bound and cutting planes to get the global optimal Bayesian network (ILPBN), and thus to get epistatic loci influencing specific phenotype traits. In order to handle large-scale of SNP loci and further to improve efficiency, we use the method of optimizing Markov blanket to reduce the number of candidate parent nodes for each node. In addition, we use α-BIC that is suitable for processing the epistatis mining to calculate the BN score. We use four properties of BN decomposable scoring functions to further reduce the number of candidate parent sets for each node. Experiment results show that ILPBN can not only process 2-locus and 3-locus epistasis mining, but also realize multi-locus epistasis detection. Finally, we compare ILPBN with several popular epistasis mining algorithms by using simulated and real Age-related macular disease (AMD) dataset. Experiment results show that ILPBN has better epistasis detection accuracy, F1-score and false positive rate in premise of ensuring the efficiency compared with other methods. Availability: Codes and dataset are available at: http://122.205.95.139/ILPBN/.

Research progress of reduced amino acid alphabets in protein analysis and prediction

•

A comprehensive summary of the literature on the reduced amino acid alphabets.

•

A systematic review of the development history of reduced amino acid alphabets.

•

Rich application cases of amino acid reduction alphabets are described in the article.

•

A detailed analysis of the properties and uses of the reduced amino acid alphabets.

Proteins are the executors of cellular physiological activities, and accurate structural and function elucidation are crucial for the refined mapping of proteins. As a feature engineering method, the reduction of amino acid composition is not only an important method for protein structure and function analysis, but also opens a broad horizon for the complex field of machine learning. Representing sequences with fewer amino acid types greatly reduces the complexity and noise of traditional feature engineering in dimension, and provides more interpretable predictive models for machine learning to capture key features. In this paper, we systematically reviewed the strategy and method studies of the reduced amino acid (RAA) alphabets, and summarized its main research in protein sequence alignment, functional classification, and prediction of structural properties, respectively. In the end, we gave a comprehensive analysis of 672 RAA alphabets from 74 reduction methods.

An enhanced binary Rat Swarm Optimizer based on local-best concepts of PSO and collaborative crossover operators for feature selection

In this paper, an enhanced binary version of the Rat Swarm Optimizer (RSO) is proposed to deal with Feature Selection (FS) problems. FS is an important data reduction step in data mining which finds the most representative features from the entire data. Many FS-based swarm intelligence algorithms have been used to tackle FS. However, the door is still open for further investigations since no FS method gives cutting-edge results for all cases. In this paper, a recent swarm intelligence metaheuristic method called RSO which is inspired by the social and hunting behavior of a group of rats is enhanced and explored for FS problems. The binary enhanced RSO is built based on three successive modifications: i) an S-shape transfer function is used to develop binary RSO algorithms; ii) the local search paradigm of particle swarm optimization is used with the iterative loop of RSO to boost its local exploitation; iii) three crossover mechanisms are used and controlled by a switch probability to improve the diversity. Based on these enhancements, three versions of RSO are produced, referred to as Binary RSO (BRSO), Binary Enhanced RSO (BERSO), and Binary Enhanced RSO with Crossover operators (BERSOC). To assess the performance of these versions, a benchmark of 24 datasets from various domains is used. The proposed methods are assessed concerning the fitness value, number of selected features, classification accuracy, specificity, sensitivity, and computational time. The best performance is achieved by BERSOC followed by BERSO and then BRSO. These proposed versions are comparatively assessed against 25 well-regarded metaheuristic methods and five filter-based approaches. The obtained results underline their superiority by producing new best results for some datasets.

Construction and Multiple Feature Classification Based on a High-Order Functional Hypernetwork on fMRI Data

Resting-state functional connectivity hypernetworks, in which multiple nodes can be connected, are an effective technique for diagnosing brain disease and performing classification research. Conventional functional hypernetworks can characterize the complex interactions within the human brain in a static form. However, an increasing body of evidence demonstrates that even in a resting state, neural activity in the brain still exhibits transient and subtle dynamics. These dynamic changes are essential for understanding the basic characteristics underlying brain organization and may correlate significantly with the pathological mechanisms of brain diseases. Therefore, considering the dynamic changes of functional connections in the resting state, we proposed methodology to construct resting state high-order functional hyper-networks (rs-HOFHNs) for patients with depression and normal subjects. Meanwhile, we also introduce a novel property (the shortest path) to extract local features with traditional local properties (cluster coefficients). A subgraph feature-based method was introduced to characterize information relating to global topology. Two features, local features and subgraph features that showed significant differences after feature selection were subjected to multi-kernel learning for feature fusion and classification. Compared with conventional hyper network models, the high-order hyper network obtained the best classification performance, 92.18%, which indicated that better classification performance can be achieved if we needed to consider multivariate interactions and the time-varying characteristics of neural interaction simultaneously when constructing a network.

EPIMUTESTR: a nearest neighbor machine learning approach to predict cancer driver genes from the evolutionary action of coding variants

Discovering rare cancer driver genes is difficult because their mutational frequency is too low for statistical detection by computational methods. EPIMUTESTR is an integrative nearest-neighbor machine learning algorithm that identifies such marginal genes by modeling the fitness of their mutations with the phylogenetic Evolutionary Action (EA) score. Over cohorts of sequenced patients from The Cancer Genome Atlas representing 33 tumor types, EPIMUTESTR detected 214 previously inferred cancer driver genes and 137 new candidates never identified computationally before of which seven genes are supported in the COSMIC Cancer Gene Census. EPIMUTESTR achieved better robustness and specificity than existing methods in a number of benchmark methods and datasets.

Binary Horse herd optimization algorithm with crossover operators for feature selection

This paper proposes a binary version of Horse herd Optimization Algorithm (HOA) to tackle Feature Selection (FS) problems. This algorithm mimics the conduct of a pack of horses when they are trying to survive. To build a Binary version of HOA, or referred to as BHOA, twofold of adjustments were made: i) Three transfer functions, namely S-shape, V-shape and U-shape, are utilized to transform the continues domain into a binary one. Four configurations of each transfer function are also well studied to yield four alternatives. ii) Three crossover operators: one-point, two-point and uniform are also suggested to ensure the efficiency of the proposed method for FS domain. The performance of the proposed fifteen BHOA versions is examined using 24 real-world FS datasets. A set of six metric measures was used to evaluate the outcome of the optimization methods: accuracy, number of features selected, fitness values, sensitivity, specificity and computational time. The best-formed version of the proposed versions is BHOA with S-shape and one-point crossover. The comparative evaluation was also accomplished against 21 state-of-the-art methods. The proposed method is able to find very competitive results where some of them are the best-recorded. Due to the viability of the proposed method, it can be further considered in other areas of machine learning.

What makes a good prediction? Feature importance and beginning to open the black box of machine learning in genetics

Genetic data have become increasingly complex within the past decade, leading researchers to pursue increasingly complex questions, such as those involving epistatic interactions and protein prediction. Traditional methods are ill-suited to answer these questions, but machine learning (ML) techniques offer an alternative solution. ML algorithms are commonly used in genetics to predict or classify subjects, but some methods evaluate which features (variables) are responsible for creating a good prediction; this is called feature importance. This is critical in genetics, as researchers are often interested in which features (e.g., SNP genotype or environmental exposure) are responsible for a good prediction. This allows for the deeper analysis beyond simple prediction, including the determination of risk factors associated with a given phenotype. Feature importance further permits the researcher to peer inside the black box of many ML algorithms to see how they work and which features are critical in informing a good prediction. This review focuses on ML methods that provide feature importance metrics for the analysis of genetic data. Five major categories of ML algorithms: k nearest neighbors, artificial neural networks, deep learning, support vector machines, and random forests are described. The review ends with a discussion of how to choose the best machine for a data set. This review will be particularly useful for genetic researchers looking to use ML methods to answer questions beyond basic prediction and classification.

Theoretical properties of distance distributions and novel metrics for nearest-neighbor feature selection

The performance of nearest-neighbor feature selection and prediction methods depends on the metric for computing neighborhoods and the distribution properties of the underlying data. Recent work to improve nearest-neighbor feature selection algorithms has focused on new neighborhood estimation methods and distance metrics. However, little attention has been given to the distributional properties of pairwise distances as a function of the metric or data type. Thus, we derive general analytical expressions for the mean and variance of pairwise distances for L_q metrics for normal and uniform random data with p attributes and m instances. The distribution moment formulas and detailed derivations provide a resource for understanding the distance properties for metrics and data types commonly used with nearest-neighbor methods, and the derivations provide the starting point for the following novel results. We use extreme value theory to derive the mean and variance for metrics that are normalized by the range of each attribute (difference of max and min). We derive analytical formulas for a new metric for genetic variants, which are categorical variables that occur in genome-wide association studies (GWAS). The genetic distance distributions account for minor allele frequency and the transition/transversion ratio. We introduce a new metric for resting-state functional MRI data (rs-fMRI) and derive its distance distribution properties. This metric is applicable to correlation-based predictors derived from time-series data. The analytical means and variances are in strong agreement with simulation results. We also use simulations to explore the sensitivity of the expected means and variances in the presence of correlation and interactions in the data. These analytical results and new metrics can be used to inform the optimization of nearest neighbor methods for a broad range of studies, including gene expression, GWAS, and fMRI data.

iBLP: An XGBoost-Based Predictor for Identifying Bioluminescent Proteins

Bioluminescent proteins (BLPs) are a class of proteins that widely distributed in many living organisms with various mechanisms of light emission including bioluminescence and chemiluminescence from luminous organisms. Bioluminescence has been commonly used in various analytical research methods of cellular processes, such as gene expression analysis, drug discovery, cellular imaging, and toxicity determination. However, the identification of bioluminescent proteins is challenging as they share poor sequence similarities among them. In this paper, we briefly reviewed the development of the computational identification of BLPs and subsequently proposed a novel predicting framework for identifying BLPs based on eXtreme gradient boosting algorithm (XGBoost) and using sequence-derived features. To train the models, we collected BLP data from bacteria, eukaryote, and archaea. Then, for getting more effective prediction models, we examined the performances of different feature extraction methods and their combinations as well as classification algorithms. Finally, based on the optimal model, a novel predictor named iBLP was constructed to identify BLPs. The robustness of iBLP has been proved by experiments on training and independent datasets. Comparison with other published method further demonstrated that the proposed method is powerful and could provide good performance for BLP identification. The webserver and software package for BLP identification are freely available at http://lin-group.cn/server/iBLP.

Scaling tree-based automated machine learning to biomedical big data with a feature set selector

Motivation

Automated machine learning (AutoML) systems are helpful data science assistants designed to scan data for novel features, select appropriate supervised learning models and optimize their parameters. For this purpose, Tree-based Pipeline Optimization Tool (TPOT) was developed using strongly typed genetic programing (GP) to recommend an optimized analysis pipeline for the data scientist’s prediction problem. However, like other AutoML systems, TPOT may reach computational resource limits when working on big data such as whole-genome expression data.

Results

We introduce two new features implemented in TPOT that helps increase the system’s scalability: Feature Set Selector (FSS) and Template. FSS provides the option to specify subsets of the features as separate datasets, assuming the signals come from one or more of these specific data subsets. FSS increases TPOT’s efficiency in application on big data by slicing the entire dataset into smaller sets of features and allowing GP to select the best subset in the final pipeline. Template enforces type constraints with strongly typed GP and enables the incorporation of FSS at the beginning of each pipeline. Consequently, FSS and Template help reduce TPOT computation time and may provide more interpretable results. Our simulations show TPOT-FSS significantly outperforms a tuned XGBoost model and standard TPOT implementation. We apply TPOT-FSS to real RNA-Seq data from a study of major depressive disorder. Independent of the previous study that identified significant association with depression severity of two modules, TPOT-FSS corroborates that one of the modules is largely predictive of the clinical diagnosis of each individual.

Availability and implementation

Detailed simulation and analysis code needed to reproduce the results in this study is available at https://github.com/lelaboratoire/tpot-fss. Implementation of the new TPOT operators is available at https://github.com/EpistasisLab/tpot.

Supplementary information

Supplementary data are available at Bioinformatics online.

Nearest-Neighbor Projected Distance Regression for Epistasis Detection in GWAS With Population Structure Correction

Nearest-neighbor Projected-Distance Regression (NPDR) is a feature selection technique that uses nearest-neighbors in high dimensional data to detect complex multivariate effects including epistasis. NPDR uses a regression formalism that allows statistical significance testing and efficient control for multiple testing. In addition, the regression formalism provides a mechanism for NPDR to adjust for population structure, which we apply to a GWAS of systemic lupus erythematosus (SLE). We also test NPDR on benchmark simulated genetic variant data with epistatic effects, main effects, imbalanced data for case-control design and continuous outcomes. NPDR identifies potential interactions in an epistasis network that influences the SLE disorder.

Nearest-neighbor Projected-Distance Regression (NPDR) for detecting network interactions with adjustments for multiple tests and confounding

SUMMARY

Machine learning feature selection methods are needed to detect complex interaction-network effects in complicated modeling scenarios in high-dimensional data, such as GWAS, gene expression, eQTL and structural/functional neuroimage studies for case-control or continuous outcomes. In addition, many machine learning methods have limited ability to address the issues of controlling false discoveries and adjusting for covariates. To address these challenges, we develop a new feature selection technique called Nearest-neighbor Projected-Distance Regression (NPDR) that calculates the importance of each predictor using generalized linear model regression of distances between nearest-neighbor pairs projected onto the predictor dimension. NPDR captures the underlying interaction structure of data using nearest-neighbors in high dimensions, handles both dichotomous and continuous outcomes and predictor data types, statistically corrects for covariates, and permits statistical inference and penalized regression. We use realistic simulations with interactions and other effects to show that NPDR has better precision-recall than standard Relief-based feature selection and random forest importance, with the additional benefit of covariate adjustment and multiple testing correction. Using RNA-Seq data from a study of major depressive disorder (MDD), we show that NPDR with covariate adjustment removes spurious associations due to confounding. We apply NPDR to eQTL data to identify potentially interacting variants that regulate transcripts associated with MDD and demonstrate NPDR's utility for GWAS and continuous outcomes.

AVAILABILITY AND IMPLEMENTATION

Available at: https://insilico.github.io/npdr/.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Consensus features nested cross-validation

SUMMARY

Feature selection can improve the accuracy of machine-learning models, but appropriate steps must be taken to avoid overfitting. Nested cross-validation (nCV) is a common approach that chooses the classification model and features to represent a given outer fold based on features that give the maximum inner-fold accuracy. Differential privacy is a related technique to avoid overfitting that uses a privacy-preserving noise mechanism to identify features that are stable between training and holdout sets.We develop consensus nested cross-validation (cnCV) that combines the idea of feature stability from differential privacy with nCV. Feature selection is applied in each inner fold and the consensus of top features across folds is used as a measure of feature stability or reliability instead of classification accuracy, which is used in standard nCV. We use simulated data with main effects, correlation and interactions to compare the classification accuracy and feature selection performance of the new cnCV with standard nCV, Elastic Net optimized by cross-validation, differential privacy and private evaporative cooling (pEC). We also compare these methods using real RNA-seq data from a study of major depressive disorder.The cnCV method has similar training and validation accuracy to nCV, but cnCV has much shorter run times because it does not construct classifiers in the inner folds. The cnCV method chooses a more parsimonious set of features with fewer false positives than nCV. The cnCV method has similar accuracy to pEC and cnCV selects stable features between folds without the need to specify a privacy threshold. We show that cnCV is an effective and efficient approach for combining feature selection with classification.

AVAILABILITY AND IMPLEMENTATION

Code available at https://github.com/insilico/cncv.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Two-Stage Classification with SIS Using a New Filter Ranking Method in High Throughput Data

Over the last decade, high dimensional data have been popularly paid attention to in bioinformatics. These data increase the likelihood of detecting the most promising novel information. However, there are limitations of high-performance computing and overfitting issues. To overcome the issues, alternative strategies need to be explored for the detection of true important features. A two-stage approach, filtering and variable selection steps, has been receiving attention. Filtering methods are divided into two categories of individual ranking and feature subset selection methods. Both have issues with the lack of consideration for joint correlation among features and computing time of an NP-hard problem. Therefore, we proposed a new filter ranking method (PF) using the elastic net penalty with sure independence screening (SIS) based on resampling technique to overcome these issues. We demonstrated that SIS-LASSO, SIS-MCP, and SIS-SCAD with the proposed filtering method achieved superior performance of not only accuracy, AUROC, and geometric mean but also true positive detection compared to those with the marginal maximum likelihood ranking method (MMLR) through extensive simulation studies. In addition, we applied it in a real application of colon and lung cancer gene expression data to investigate the classification performance and power of detecting true genes associated with colon and lung cancer.

Transition-transversion encoding and genetic relationship metric in ReliefF feature selection improves pathway enrichment in GWAS

BACKGROUND

ReliefF is a nearest-neighbor based feature selection algorithm that efficiently detects variants that are important due to statistical interactions or epistasis. For categorical predictors, like genotypes, the standard metric used in ReliefF has been a simple (binary) mismatch difference. In this study, we develop new metrics of varying complexity that incorporate allele sharing, adjustment for allele frequency heterogeneity via the genetic relationship matrix (GRM), and physicochemical differences of variants via a new transition/transversion encoding.

METHODS

We introduce a new two-dimensional transition/transversion genotype encoding for ReliefF, and we implement three ReliefF attribute metrics: 1.) genotype mismatch (GM), which is the ReliefF standard, 2.) allele mismatch (AM), which accounts for heterozygous differences and has not been used previously in ReliefF, and 3.) the new transition/transversion metric. We incorporate these attribute metrics into the ReliefF nearest neighbor calculation with a Manhattan metric, and we introduce GRM as a new ReliefF nearest-neighbor metric to adjust for allele frequency heterogeneity.

RESULTS

We apply ReliefF with each metric to a GWAS of major depressive disorder and compare the detection of genes in pathways implicated in depression, including Axon Guidance, Neuronal System, and G Protein-Coupled Receptor Signaling. We also compare with detection by Random Forest and Lasso as well as random/null selection to assess pathway size bias.

CONCLUSIONS

Our results suggest that using more genetically motivated encodings, such as transition/transversion, and metrics that adjust for allele frequency heterogeneity, such as GRM, lead to ReliefF attribute scores with improved pathway enrichment.