Elastic SCAD as a novel penalization method for SVM classification tasks in high-dimensional data

The impact of temporal lobe epilepsy surgery on picture naming and its relationship to network metric change

BACKGROUND

Anterior temporal lobe resection (ATLR) is a successful treatment for medically-refractory temporal lobe epilepsy (TLE). In the language-dominant hemisphere, 30%- 50% of individuals experience a naming decline which can impact upon daily life. Measures of structural networks are associated with language performance pre-operatively. It is unclear if analysis of network measures may predict post-operative decline.

METHODS

White matter fibre tractography was performed on preoperative diffusion MRI of 44 left lateralised and left resection individuals with TLE to reconstruct the preoperative structural network. Resection masks, drawn on co-registered pre- and post-operative T1-weighted MRI scans, were used as exclusion regions on pre-operative tractography to estimate the post-operative network. Changes in graph theory metrics, cortical strength, betweenness centrality, and clustering coefficient were generated by comparing the estimated pre- and post-operative networks. These were thresholded based on the presence of the connection in each patient, ranging from 75% to 100% in steps of 5%. The average graph theory metric across thresholds was taken. We incorporated leave-one-out cross-validation with smoothly clipped absolute deviation (SCAD) least absolute shrinkage and selection operator (LASSO) feature selection and a support vector classifier to assess graph theory metrics on picture naming decline. Picture naming was assessed via the Graded Naming Test preoperatively and at 3 and 12 months post-operatively and the outcome was classified using the reliable change index (RCI) to identify clinically significant decline. The best feature combination and model was selected using the area under the curve (AUC). The sensitivity, specificity and F1-score were also reported. Permutation testing was performed to assess the machine learning model and selected regions difference significance.

RESULTS

A combination of clinical and graph theory metrics were able to classify outcome of picture naming at 3 months with an AUC of 0.84. At 12 months, change in strength to cortical regions was best able to correctly classify outcome with an AUC of 0.86. Longitudinal analysis revealed that betweenness centrality was the best metric to identify patients who declined at 3 months, who will then continue to experience decline from 3 to 12 months. Both models were significantly higher AUC values than a random classifier.

CONCLUSION

Our results suggest that inferred changes of network integrity were able to correctly classify picture naming decline after ATLR. These measures may be used to prospectively to identify patients who are at risk of picture naming decline after surgery and could potentially be utilised to assist tailoring the resection in order to prevent this decline.

A novel meta-analysis based on data augmentation and elastic data shared lasso regularization for gene expression

Background

Gene expression analysis can provide useful information for analyzing complex biological mechanisms. However, many reported findings are unrepeatable due to small sample sizes relative to a large number of genes and the low signal-to-noise ratios of most gene expression datasets.

Results

Meta-analysis of multi-data sets is an efficient method for tackling the above problem. To improve the performance of meta-analysis, we propose a novel meta-analysis framework. It consists of two parts: (1) a novel data augmentation strategy. Various cross-platform normalization methods exist, which can preserve original biological information of gene expression datasets from different angles and add different “perturbations” to the dataset. Using such perturbation, we provide a feasible means for gene expression data augmentation; (2) elastic data shared lasso (DSL-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\varvec{L}}}_{\mathbf{2}}$$\end{document}L2). The DSL-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{L}}_{\mathbf{2}}$$\end{document}L2 method spans the continuum between individual models for each dataset and one model for all datasets. It also overcomes the shortcomings of the data shared lasso method when dealing with highly correlated features. Comprehensive simulation experiment results show that the proposed method has high prediction and gene selection performance. We then apply the proposed method to non-small cell lung cancer (NSCLC) blood gene expression data in order to identify key tumor-related genes. The outcomes of our experiment indicate that the method could be used for identifying a set of robust disease-related gene signatures that may be used for NSCLC early diagnosis or prognosis or even targeting.

Conclusion

We propose a novel and effective meta-analysis method for biological research, extrapolating and integrating information from multiple gene expression datasets.

Myocardial Function Prediction After Coronary Artery Bypass Grafting Using MRI Radiomic Features and Machine Learning Algorithms

The main aim of the present study was to predict myocardial function improvement in cardiac MR (LGE-CMR) images in patients after coronary artery bypass grafting (CABG) using radiomics and machine learning algorithms. Altogether, 43 patients who had visible scars on short-axis LGE-CMR images and were candidates for CABG surgery were selected and enrolled in this study. MR imaging was performed preoperatively using a 1.5-T MRI scanner. All images were segmented by two expert radiologists (in consensus). Prior to extraction of radiomics features, all MR images were resampled to an isotropic voxel size of 1.8 × 1.8 × 1.8 mm³. Subsequently, intensities were quantized to 64 discretized gray levels and a total of 93 features were extracted. The applied algorithms included a smoothly clipped absolute deviation (SCAD)-penalized support vector machine (SVM) and the recursive partitioning (RP) algorithm as a robust classifier for binary classification in this high-dimensional and non-sparse data. All models were validated with repeated fivefold cross-validation and 10,000 bootstrapping resamples. Ten and seven features were selected with SCAD-penalized SVM and RP algorithm, respectively, for CABG responder/non-responder classification. Considering univariate analysis, the GLSZM gray-level non-uniformity-normalized feature achieved the best performance (AUC: 0.62, 95% CI: 0.53-0.76) with SCAD-penalized SVM. Regarding multivariable modeling, SCAD-penalized SVM obtained an AUC of 0.784 (95% CI: 0.64-0.92), whereas the RP algorithm achieved an AUC of 0.654 (95% CI: 0.50-0.82). In conclusion, different radiomics texture features alone or combined in multivariate analysis using machine learning algorithms provide prognostic information regarding myocardial function in patients after CABG.

Communication-Efficient Distributed Learning for High-Dimensional Support Vector Machines

Distributed learning has received increasing attention in recent years and is a special need for the era of big data. For a support vector machine (SVM), a powerful binary classification tool, we proposed a novel efficient distributed sparse learning algorithm, the communication-efficient surrogate likelihood support vector machine (CSLSVM), in high-dimensions with convex or nonconvex penalties, based on a communication-efficient surrogate likelihood (CSL) framework. We extended the CSL for distributed SVMs without the need to smooth the hinge loss or the gradient of the loss. For a CSLSVM with lasso penalty, we proved that its estimator could achieve a near-oracle property for l1 penalized SVM estimators on whole datasets. For a CSLSVM with smoothly clipped absolute deviation penalty, we showed that its estimator enjoyed the oracle property, and that it used local linear approximation (LLA) to solve the optimization problem. Furthermore, we showed that the LLA was guaranteed to converge to the oracle estimator, even in our distributed framework and the ultrahigh-dimensional setting, if an appropriate initial estimator was available. The proposed approach is highly competitive with the centralized method within a few rounds of communications. Numerical experiments provided supportive evidence.

Prediction of Decline in Global Cognitive Function Using Machine Learning with Feature Ranking of Gait and Physical Fitness Outcomes in Older Adults

Gait and physical fitness are related to cognitive function. A decrease in motor function and physical fitness can serve as an indicator of declining global cognitive function in older adults. This study aims to use machine learning (ML) to identify important features of gait and physical fitness to predict a decline in global cognitive function in older adults. A total of three hundred and six participants aged seventy-five years or older were included in the study, and their gait performance at various speeds and physical fitness were evaluated. Eight ML models were applied to data ranked by the p-value (LP) of linear regression and the importance gain (XI) of XGboost. Five optimal features were selected using elastic net on the LP data for men, and twenty optimal features were selected using support vector machine on the XI data for women. Thus, the important features for predicting a potential decline in global cognitive function in older adults were successfully identified herein. The proposed ML approach could inspire future studies on the early detection and prevention of cognitive function decline in older adults.

Seeking diagnostic and prognostic biomarkers for childhood bacterial pneumonia in sub-Saharan Africa: study protocol for an observational study

INTRODUCTION

Clinically diagnosed pneumonia in children is a leading cause of paediatric hospitalisation and mortality. The aetiology is usually bacterial or viral, but malaria can cause a syndrome indistinguishable from clinical pneumonia. There is no method with high sensitivity to detect a bacterial infection in these patients and, as result, antibiotics are frequently overprescribed. Conversely, unrecognised concomitant bacterial infection in patients with malarial infections occur with omission of antibiotic therapy from patients with bacterial infections. Previously, we identified two combinations of blood proteins with 96% sensitivity and 86% specificity for detecting bacterial disease. The current project aimed to validate and improve these combinations by evaluating additional biomarkers in paediatric patients with clinical pneumonia. Our goal was to describe combinations of a limited number of proteins with high sensitivity and specificity for bacterial infection to be incorporated in future point-of-care tests. Furthermore, we seek to explore signatures to prognosticate clinical pneumonia.

METHODS AND ANALYSIS

Patients (n=900) aged 2-59 months presenting with clinical pneumonia at two Gambian hospitals will be enrolled and classified according to criteria for definitive bacterial aetiology (based on microbiological tests and chest radiographs). We will measure proteins at admission using Luminex-based immunoassays in 90 children with definitive and 160 with probable bacterial aetiology, and 160 children classified according to the prognosis of their disease. Previously identified diagnostic signatures will be assessed through accuracy measures. Moreover, we will seek new diagnostic and prognostic signatures through machine learning methods, including support vector machine, penalised regression and classification trees.

ETHICS AND DISSEMINATION

Ethics approval has been obtained from the Gambia Government/Medical Research Council Unit The Gambia Joint Ethics Committee (protocol 1616) and the institutional review board of Boston University Medical Centre (STUDY00000958). Study results will be disseminated to the staff of the study hospitals, in scientific seminars and meetings, and in publications.

TRIAL REGISTRATION NUMBER

H-38462.

Distinguishing Rectal Cancer from Colon Cancer Based on the Support Vector Machine Method and RNA-sequencing Data

Colorectal cancer (CRC) is the third most commonly diagnosed cancer worldwide. Several studies have indicated that rectal cancer is significantly different from colon cancer in terms of treatment, prognosis, and metastasis. Recently, the differential mRNA expression of colon cancer and rectal cancer has received a great deal of attention. The current study aimed to identify significant differences between colon cancer and rectal cancer based on RNA sequencing (RNA-seq) data via support vector machines (SVM). Here, 393 CRC samples from the The Cancer Genome Atlas (TCGA) database were investigated, including 298 patients with colon cancer and 95 with rectal cancer. Following the random forest (RF) analysis of the mRNA expression data, 96 genes such as HOXB13, PRAC, and BCLAF1 were identified and utilized to build the SVM classification model with the Leave-One-Out Cross-validation (LOOCV) algorithm. In the training (n=196) and the validation cohorts (n=197), the accuracy (82.1 % and 82.2 %, respectively) and the AUC (0.87 and 0.91, respectively) indicated that the established optimal SVM classification model distinguished colon cancer from rectal cancer reasonably. However, additional experiments are required to validate the predicted gene expression levels and functions.

EnACP: An Ensemble Learning Model for Identification of Anticancer Peptides

As cancer remains one of the main threats of human life, developing efficient cancer treatments is urgent. Anticancer peptides, which could overcome the significant side effects and poor results of traditional cancer treatments, have become a new potential alternative these years. However, identifying anticancer peptides by experimental methods is time consuming and resource consuming, it is of great significance to develop effective computational tools to quickly and accurately identify potential anticancer peptides from amino acid sequences. For most current computational methods, feature representation plays a key role in their final successes. This study proposes a novel fast and accurate approach to identify anticancer peptides using diversified feature representations and ensemble learning method. For the feature representations, the information is encoded from multidimensional feature spaces, including sequence composition, sequence-order, physicochemical properties, etc. In order to better model the potential relationships of peptides, multiple ensemble classifiers, LightGBMs, are applied to detect the different feature sets at first. Then the obtained multiple outputs are used as inputs of the support vector machine classifier, which effectively identifies anticancer peptides. Experimental results on cross validation and independent test sets demonstrate that our method can achieve better or comparable performances compared with other state-of-the-art methods.

Bayesian Non-linear Support Vector Machine for High-Dimensional Data with Incorporation of Graph Information on Features

Support vector machine (SVM) is a popular classification method for analysis of high dimensional data such as genomics data. Recently a number of linear SVM methods have been developed to achieve feature selection through either frequentist regularization or Bayesian shrinkage, but the linear assumption may not be plausible for many real applications. In addition, recent work has demonstrated that incorporating known biological knowledge, such as those from functional genomics, into the statistical analysis of genomic data offers great promise of improved predictive accuracy and feature selection. Such biological knowledge can often be represented by graphs. In this article, we propose a novel knowledge-guided nonlinear Bayesian SVM approach for analysis of high-dimensional data. Our model uses graph information that represents the relationship among the features to guide feature selection. To achieve knowledge-guided feature selection, we assign an Ising prior to the indicators representing inclusion/exclusion of the features in the model. An efficient MCMC algorithm is developed for posterior inference. The performance of our method is evaluated and compared with several penalized linear SVM and the standard kernel SVM method in terms of prediction and feature selection in extensive simulation studies. Also, analyses of genomic data from a cancer study show that our method yields a more accurate prediction model for patient survival and reveals biologically more meaningful results than the existing methods.

An ensemble forecast model of dengue in Guangzhou, China using climate and social media surveillance data

BACKGROUND

China experienced an unprecedented outbreak of dengue in 2014, and the number of dengue cases reached the highest level over the past 25 years. There is a significant delay in the release of official case count data, and our ability to timely track the timing and magnitude of local outbreaks of dengue remains limited.

MATERIAL AND METHODS

We developed an ensemble penalized regression algorithm (EPRA) for initializing near-real time forecasts of the dengue epidemic trajectory by integrating different penalties (LASSO, Ridge, Elastic Net, SCAD and MCP) with the techniques of iteratively sampling and model averaging. Multiple streams of near-real time data including dengue-related Baidu searches, Sina Weibo posts, and climatic conditions with historical dengue incidence were used. We compared the predictive power of the EPRA with the alternates, penalized regression models using single penalties, to retrospectively forecast weekly dengue incidence and detect outbreak occurrence defined using different cutoffs, during the periods of 2011-2016 in Guangzhou, south China.

RESULTS

The EPRA showed the best or at least comparable performance for 1-, 2-week ahead out-of-sample and leave-one-out cross validation forecasts. The findings indicate that skillful near-real time forecasts of dengue and confidence in those predictions can be made. For detecting dengue outbreaks, the EPRA predicted periods of high incidence of dengue more accurately than the alternates.

CONCLUSION

This study developed a statistically rigorous approach for near-real time forecast of dengue in China. The EPRA provides skillful forecasts and can be used as timely and complementary ways to assess dengue dynamics, which will help to design interventions to mitigate dengue transmission.

Knowledge-Guided Bayesian Support Vector Machine for High-Dimensional Data with Application to Analysis of Genomics Data

Support vector machine (SVM) is a popular classification method for the analysis of wide range of data including big data. Many SVM methods with feature selection have been developed under frequentist regularization or Bayesian shrinkage frameworks. On the other hand, the importance of incorporating a priori known biological knowledge, such as gene pathway information which stems from the gene regulatory network, into the statistical analysis of genomic data has been recognized in recent years. In this article, we propose a new Bayesian SVM approach that enables the feature selection to be guided by the knowledge on the graphical structure among predictors. The proposed method uses the spike-and-slab prior for feature selection, combined with the Ising prior that encourages group-wise selection of the predictors adjacent to each other on the known graph. Gibbs sampling algorithm is used for Bayesian inference. The performance of our method is evaluated and compared with existing SVM methods in terms of prediction and feature selection in extensive simulation settings. In addition, our method is illustrated in the analysis of genomic data from a cancer study, demonstrating its advantage in generating biologically meaningful results and identifying potentially important features.

SVM-RFE: selection and visualization of the most relevant features through non-linear kernels

Background

Support vector machines (SVM) are a powerful tool to analyze data with a number of predictors approximately equal or larger than the number of observations. However, originally, application of SVM to analyze biomedical data was limited because SVM was not designed to evaluate importance of predictor variables. Creating predictor models based on only the most relevant variables is essential in biomedical research. Currently, substantial work has been done to allow assessment of variable importance in SVM models but this work has focused on SVM implemented with linear kernels. The power of SVM as a prediction model is associated with the flexibility generated by use of non-linear kernels. Moreover, SVM has been extended to model survival outcomes. This paper extends the Recursive Feature Elimination (RFE) algorithm by proposing three approaches to rank variables based on non-linear SVM and SVM for survival analysis.

Results

The proposed algorithms allows visualization of each one the RFE iterations, and hence, identification of the most relevant predictors of the response variable. Using simulation studies based on time-to-event outcomes and three real datasets, we evaluate the three methods, based on pseudo-samples and kernel principal component analysis, and compare them with the original SVM-RFE algorithm for non-linear kernels. The three algorithms we proposed performed generally better than the gold standard RFE for non-linear kernels, when comparing the truly most relevant variables with the variable ranks produced by each algorithm in simulation studies. Generally, the RFE-pseudo-samples outperformed the other three methods, even when variables were assumed to be correlated in all tested scenarios.

Conclusions

The proposed approaches can be implemented with accuracy to select variables and assess direction and strength of associations in analysis of biomedical data using SVM for categorical or time-to-event responses. Conducting variable selection and interpreting direction and strength of associations between predictors and outcomes with the proposed approaches, particularly with the RFE-pseudo-samples approach can be implemented with accuracy when analyzing biomedical data. These approaches, perform better than the classical RFE of Guyon for realistic scenarios about the structure of biomedical data.

Electronic supplementary material

The online version of this article (10.1186/s12859-018-2451-4) contains supplementary material, which is available to authorized users.

Hybrid L1/2 + 2 method for gene selection in the Cox proportional hazards model

BACKGROUND AND OBJECTIVE

An important issue in genomic research is to identify the significant genes that related to survival from tens of thousands of genes. Although Cox proportional hazards model is a conventional survival analysis method, it does not induce the gene selection.

METHODS

In this paper, we extend the hybrid L_{1/2  + 2} regularization (HLR) idea to the censored survival situation, a new edition of sparse Cox model based on the HLR method has been proposed. We develop two algorithms for solving the HLR penalized Cox model; one is the coordinate descent algorithm with HLR thresholding operator, the other is the weight iteration method.

RESULTS

The proposed method was tested on six public mRNA data sets of serval kinds of cancers, AML, Breast cancer, Pancreatic cancer, DLBCL and Melanoma. The test results indicate that the method identified a small subset of genes but essential while giving best or equivalent predictive performance, as compared to some popular methods.

CONCLUSIONS

The results of empirical and simulations imply that the proposed strategy is highly competitive in studying high dimensional survival data among several state-of-the-art methods.

Finding disagreement pathway signatures and constructing an ensemble model for cancer classification

Cancer classification based on molecular level is a relatively routine research procedure with advances in high-throughput molecular profiling techniques. However, the number of genes typically far exceeds the number of the sample size in gene expression studies. The existing gene selection methods are almost based on statistics and machine learning, overlooking relevant biological principles or knowledge while working with biological data. Here, we propose a robust ensemble learning paradigm, which incorporates multiple pathways information, to predict cancer classification. We compare the proposed method with other methods, such as Elastic SCAD and PPDMF, and estimate the classification performance. The results show that the proposed method has the higher performances on most metrics and robust performance. We further investigate the biological mechanism of the ensemble feature genes. The results demonstrate that the ensemble feature genes are associated with drug targets/clinically-relevant cancer. In addition, some core biological pathways and biological process underlying clinically-relevant phenotypes are identified by function annotation. Overall, our research can provide a new perspective for the further study of molecular activities and manifestations of cancer.

A QSAR classification model for neuraminidase inhibitors of influenza A viruses (H1N1) based on weighted penalized support vector machine

Descriptor selection is a procedure widely used in chemometrics. The aim is to select the best subset of descriptors relevant to the quantitative structure-activity relationship (QSAR) study being considered. In this paper, a new descriptor selection method for the QSAR classification model is proposed by adding a new weight inside L1-norm. The experimental results from classifying the neuraminidase inhibitors of influenza A viruses (H1N1) demonstrate that the proposed method in the QSAR classification model performs effectively and competitively compared with other existing penalized methods in terms of classification performance and the number of selected descriptors.

Survival Prediction and Feature Selection in Patients with Breast Cancer Using Support Vector Regression

The Support Vector Regression (SVR) model has been broadly used for response prediction. However, few researchers have used SVR for survival analysis. In this study, a new SVR model is proposed and SVR with different kernels and the traditional Cox model are trained. The models are compared based on different performance measures. We also select the best subset of features using three feature selection methods: combination of SVR and statistical tests, univariate feature selection based on concordance index, and recursive feature elimination. The evaluations are performed using available medical datasets and also a Breast Cancer (BC) dataset consisting of 573 patients who visited the Oncology Clinic of Hamadan province in Iran. Results show that, for the BC dataset, survival time can be predicted more accurately by linear SVR than nonlinear SVR. Based on the three feature selection methods, metastasis status, progesterone receptor status, and human epidermal growth factor receptor 2 status are the best features associated to survival. Also, according to the obtained results, performance of linear and nonlinear kernels is comparable. The proposed SVR model performs similar to or slightly better than other models. Also, SVR performs similar to or better than Cox when all features are included in model.

Cancers Screening in an Asymptomatic Population by Using Multiple Tumour Markers

BACKGROUND

Analytic measurement of serum tumour markers is one of commonly used methods for cancer risk management in certain areas of the world (e.g. Taiwan). Recently, cancer screening based on multiple serum tumour markers has been frequently discussed. However, the risk-benefit outcomes appear to be unfavourable for patients because of the low sensitivity and specificity. In this study, cancer screening models based on multiple serum tumour markers were designed using machine learning methods, namely support vector machine (SVM), k-nearest neighbour (KNN), and logistic regression, to improve the screening performance for multiple cancers in a large asymptomatic population.

METHODS

AFP, CEA, CA19-9, CYFRA21-1, and SCC were determined for 20 696 eligible individuals. PSA was measured in men and CA15-3 and CA125 in women. A variable selection process was applied to select robust variables from these serum tumour markers to design cancer detection models. The sensitivity, specificity, positive predictive value (PPV), negative predictive value, area under the curve, and Youden index of the models based on single tumour markers, combined test, and machine learning methods were compared. Moreover, relative risk reduction, absolute risk reduction (ARR), and absolute risk increase (ARI) were evaluated.

RESULTS

To design cancer detection models using machine learning methods, CYFRA21-1 and SCC were selected for women, and all tumour markers were selected for men. SVM and KNN models significantly outperformed the single tumour markers and the combined test for men. All 3 studied machine learning methods outperformed single tumour markers and the combined test for women. For either men or women, the ARRs were between 0.003-0.008; the ARIs were between 0.119-0.306.

CONCLUSION

Machine learning methods outperformed the combined test in analysing multiple tumour markers for cancer detection. However, cancer screening based solely on the application of multiple tumour markers remains unfavourable because of the inadequate PPV, ARR, and ARI, even when machine learning methods were incorporated into the analysis.

Feature Selection and Cancer Classification via Sparse Logistic Regression with the Hybrid L1/2 +2 Regularization

Cancer classification and feature (gene) selection plays an important role in knowledge discovery in genomic data. Although logistic regression is one of the most popular classification methods, it does not induce feature selection. In this paper, we presented a new hybrid L_{1/2 +2} regularization (HLR) function, a linear combination of L_1/2 and L₂ penalties, to select the relevant gene in the logistic regression. The HLR approach inherits some fascinating characteristics from L_1/2 (sparsity) and L₂ (grouping effect where highly correlated variables are in or out a model together) penalties. We also proposed a novel univariate HLR thresholding approach to update the estimated coefficients and developed the coordinate descent algorithm for the HLR penalized logistic regression model. The empirical results and simulations indicate that the proposed method is highly competitive amongst several state-of-the-art methods.

Classification of radiology reports for falls in an HIV study cohort

OBJECTIVE

To identify patients in a human immunodeficiency virus (HIV) study cohort who have fallen by applying supervised machine learning methods to radiology reports of the cohort.

METHODS

We used the Veterans Aging Cohort Study Virtual Cohort (VACS-VC), an electronic health record-based cohort of 146 530 veterans for whom radiology reports were available (N=2 977 739). We created a reference standard of radiology reports, represented each report by a feature set of words and Unified Medical Language System concepts, and then developed several support vector machine (SVM) classifiers for falls. We compared mutual information (MI) ranking and embedded feature selection approaches. The SVM classifier with MI feature selection was chosen to classify all radiology reports in VACS-VC.

RESULTS

Our SVM classifier with MI feature selection achieved an area under the curve score of 97.04 on the test set. When applied to all the radiology reports in VACS-VC, 80 416 of these reports were classified as positive for a fall. Of these, 11 484 were associated with a fall-related external cause of injury code (E-code) and 68 932 were not, corresponding to 29 280 patients with potential fall-related injuries who could not have been found using E-codes.

DISCUSSION

Feature selection was crucial to improving the classifier's performance. Feature selection with MI allowed us to select the number of discriminative features to use for classification, in contrast to the embedded feature selection method, in which the number of features is chosen automatically.

CONCLUSION

Machine learning is an effective method of identifying patients who have suffered a fall. The development of this classifier supplements the clinical researcher's toolkit and reduces dependence on under-coded structured electronic health record data.

A Consistent Information Criterion for Support Vector Machines in Diverging Model Spaces

Information criteria have been popularly used in model selection and proved to possess nice theoretical properties. For classification, Claeskens et al. (2008) proposed support vector machine information criterion for feature selection and provided encouraging numerical evidence. Yet no theoretical justification was given there. This work aims to fill the gap and to provide some theoretical justifications for support vector machine information criterion in both fixed and diverging model spaces. We first derive a uniform convergence rate for the support vector machine solution and then show that a modification of the support vector machine information criterion achieves model selection consistency even when the number of features diverges at an exponential rate of the sample size. This consistency result can be further applied to selecting the optimal tuning parameter for various penalized support vector machine methods. Finite-sample performance of the proposed information criterion is investigated using Monte Carlo studies and one real-world gene selection problem.

Proposal of supervised data analysis strategy of plasma miRNAs from hybridisation array data with an application to assess hemolysis-related deregulation

Background

Plasma miRNAs have the potential as cancer biomarkers but no consolidated guidelines for data mining in this field are available. The purpose of the study was to apply a supervised data analysis strategy in a context where prior knowledge is available, i.e., that of hemolysis-related miRNAs deregulation, so as to compare our results with existing evidence.

Results

We developed a structured strategy with innovative applications of existing bioinformatics methods for supervised analyses including: 1) the combination of two statistical (t- and Anderson-Darling) test results to detect miRNAs with significant fold change or general distributional differences in class comparison, which could reveal hidden differential biological processes worth to be considered for building predictive tools; 2) a bootstrap selection procedure together with machine learning techniques in class prediction to guarantee the transferability of results and explore the interconnections among the selected miRNAs, which is important for highlighting their inherent biological dependences. The strategy was applied to develop a classifier for discriminating between hemolyzed and not hemolyzed plasma samples, defined according to a recently published hemolysis score. We identified five miRNAs with increased expression in hemolyzed plasma samples (miR-486-5p, miR-92a, miR-451, miR-16, miR-22).

Conclusions

We identified four miRNAs previously reported in the literature as hemolysis related together with a new one (miR-22).which needs further investigations. Our findings confirm the validity of the proposed strategy and, in parallel, the hemolysis score capability to be used as pre-analytic hemolysis detector. R codes for implementing the approaches are provided.

Electronic supplementary material

The online version of this article (doi:10.1186/s12859-015-0820-9) contains supplementary material, which is available to authorized users.

Sparse Modeling Reveals miRNA Signatures for Diagnostics of Inflammatory Bowel Disease

The diagnosis of inflammatory bowel disease (IBD) still remains a clinical challenge and the most accurate diagnostic procedure is a combination of clinical tests including invasive endoscopy. In this study we evaluated whether systematic miRNA expression profiling, in conjunction with machine learning techniques, is suitable as a non-invasive test for the major IBD phenotypes (Crohn's disease (CD) and ulcerative colitis (UC)). Based on microarray technology, expression levels of 863 miRNAs were determined for whole blood samples from 40 CD and 36 UC patients and compared to data from 38 healthy controls (HC). To further discriminate between disease-specific and general inflammation we included miRNA expression data from other inflammatory diseases (inflammation controls (IC): 24 chronic obstructive pulmonary disease (COPD), 23 multiple sclerosis, 38 pancreatitis and 45 sarcoidosis cases) as well as 70 healthy controls from previous studies. Classification problems considering 2, 3 or 4 groups were solved using different types of penalized support vector machines (SVMs). The resulting models were assessed regarding sparsity and performance and a subset was selected for further investigation. Measured by the area under the ROC curve (AUC) the corresponding median holdout-validated accuracy was estimated as ranging from 0.75 to 1.00 (including IC) and 0.89 to 0.98 (excluding IC), respectively. In combination, the corresponding models provide tools for the distinction of CD and UC as well as CD, UC and HC with expected classification error rates of 3.1 and 3.3%, respectively. These results were obtained by incorporating not more than 16 distinct miRNAs. Validated target genes of these miRNAs have been previously described as being related to IBD. For others we observed significant enrichment for IBD susceptibility loci identified in earlier GWAS. These results suggest that the proposed miRNA signature is of relevance for the etiology of IBD. Its diagnostic value, however, should be further evaluated in large, independent, clinically well characterized cohorts.

Differential scanning calorimetry as a complementary diagnostic tool for the evaluation of biological samples

BACKGROUND

Differential scanning calorimetry (DSC) is a tool for measuring the thermal stability profiles of complex molecular interactions in biological fluids. DSC profiles (thermograms) of biofluids provide specific signatures which are being utilized as a new diagnostic approach for characterizing disease but the development of these approaches is still in its infancy.

METHODS

This article evaluates several approaches for the analysis of thermograms which could increase the utility of DSC for clinical application. Thermograms were analyzed using localized thermogram features and principal components (PCs). The performance of these methods was evaluated alongside six models for the classification of a data set comprised of 300 systemic lupus erythematosus (SLE) patients and 300 control subjects obtained from the Lupus Family Registry and Repository (LFRR).

RESULTS

Classification performance was substantially higher using the penalized algorithms relative to localized features/PCs alone. The models were grouped into two sets, the first having smoother solution vectors but lower classification accuracies than the second with seemingly noisier solution vectors.

CONCLUSIONS

Coupling thermogram technology with modern classification algorithms provides a powerful diagnostic approach for analysis of biological samples. The solution vectors from the models may reflect important information from the thermogram profiles for discriminating between clinical groups.

GENERAL SIGNIFICANCE

DSC thermograms show sensitivity to changes in the bulk plasma proteome that correlate with clinical status. To move this technology towards clinical application the development of new approaches is needed to extract discriminatory parameters from DSC profiles for the comparison and diagnostic classification of patients.

Secondary electrospray ionization-mass spectrometry and a novel statistical bioinformatic approach identifies a cancer-related profile in exhaled breath of breast cancer patients: a pilot study

Breath analysis represents a new frontier in medical diagnosis and a powerful tool for cancer biomarker discovery due to the recent development of analytical platforms for the detection and identification of human exhaled volatile compounds. Statistical and bioinformatic tools may represent an effective complement to the technical and instrumental enhancements needed to fully exploit clinical applications of breath analysis. Our exploratory study in a cohort of 14 breast cancer patients and 11 healthy volunteers used secondary electrospray ionization-mass spectrometry (SESI-MS) to detect a cancer-related volatile profile. SESI-MS full-scan spectra were acquired in a range of 40-350 mass-to-charge ratio (m/z), converted to matrix data and analyzed using a procedure integrating data pre-processing for quality control, and a two-step class prediction based on machine-learning techniques, including a robust feature selection, and a classifier development with internal validation. MS spectra from exhaled breath showed an individual-specific breath profile and high reciprocal homogeneity among samples, with strong agreement among technical replicates, suggesting a robust responsiveness of SESI-MS. Supervised analysis of breath data identified a support vector machine (SVM) model including 8 features corresponding to m/z 106, 126, 147, 78, 148, 52, 128, 315 and able to discriminate exhaled breath from breast cancer patients from that of healthy individuals, with sensitivity and specificity above 0.9.Our data highlight the significance of SESI-MS as an analytical technique for clinical studies of breath analysis and provide evidence that our noninvasive strategy detects volatile signatures that may support existing technologies to diagnose breast cancer.

Variable Selection for Support Vector Machines in Moderately High Dimensions

The support vector machine (SVM) is a powerful binary classification tool with high accuracy and great flexibility. It has achieved great success, but its performance can be seriously impaired if many redundant covariates are included. Some efforts have been devoted to studying variable selection for SVMs, but asymptotic properties, such as variable selection consistency, are largely unknown when the number of predictors diverges to infinity. In this work, we establish a unified theory for a general class of nonconvex penalized SVMs. We first prove that in ultra-high dimensions, there exists one local minimizer to the objective function of nonconvex penalized SVMs possessing the desired oracle property. We further address the problem of nonunique local minimizers by showing that the local linear approximation algorithm is guaranteed to converge to the oracle estimator even in the ultra-high dimensional setting if an appropriate initial estimator is available. This condition on initial estimator is verified to be automatically valid as long as the dimensions are moderately high. Numerical examples provide supportive evidence.

Prognostic gene signatures for patient stratification in breast cancer: accuracy, stability and interpretability of gene selection approaches using prior knowledge on protein-protein interactions

Background

Stratification of patients according to their clinical prognosis is a desirable goal in cancer treatment in order to achieve a better personalized medicine. Reliable predictions on the basis of gene signatures could support medical doctors on selecting the right therapeutic strategy. However, during the last years the low reproducibility of many published gene signatures has been criticized. It has been suggested that incorporation of network or pathway information into prognostic biomarker discovery could improve prediction performance. In the meanwhile a large number of different approaches have been suggested for the same purpose.

Methods

We found that on average incorporation of pathway information or protein interaction data did not significantly enhance prediction performance, but indeed greatly interpretability of gene signatures. Some methods (specifically network-based SVMs) could greatly enhance gene selection stability, but revealed only a comparably low prediction accuracy, whereas Reweighted Recursive Feature Elimination (RRFE) and average pathway expression led to very clearly interpretable signatures. In addition, average pathway expression, together with elastic net SVMs, showed the highest prediction performance here.

Results

The results indicated that no single algorithm to perform best with respect to all three categories in our study. Incorporating network of prior knowledge into gene selection methods in general did not significantly improve classification accuracy, but greatly interpretability of gene signatures compared to classical algorithms.

Biomarker gene signature discovery integrating network knowledge

Discovery of prognostic and diagnostic biomarker gene signatures for diseases, such as cancer, is seen as a major step towards a better personalized medicine. During the last decade various methods, mainly coming from the machine learning or statistical domain, have been proposed for that purpose. However, one important obstacle for making gene signatures a standard tool in clinical diagnosis is the typical low reproducibility of these signatures combined with the difficulty to achieve a clear biological interpretation. For that purpose in the last years there has been a growing interest in approaches that try to integrate information from molecular interaction networks. Here we review the current state of research in this field by giving an overview about so-far proposed approaches.