Feature selection and survival modeling in The Cancer Genome Atlas

Discovery of pathway-independent protein signatures associated with clinical outcome in human cancer cohorts

Proteomic data provide a direct readout of protein function, thus constituting an information-rich resource for prognostic and predictive modeling. However, protein array data may not fully capture pathway activity due to the limited number of molecules and incomplete pathway coverage compared to other high-throughput technologies. For the present study, our aim was to improve clinical outcome prediction compared to published pathway-dependent prognostic signatures for The Cancer Genome Atlas (TCGA) cohorts using the least absolute shrinkage and selection operator (LASSO). RPPA data is particularly well-suited to the LASSO due to the relatively low number of predictors compared to larger genomic data matrices. Our approach selected predictors regardless of their pathway membership and optimally combined their RPPA measurements into a weighted risk score. Performance was assessed and compared to that of the published signatures using two unbiased approaches: 1) 10 iterations of threefold cross-validation for unbiased estimation of hazard ratio and difference in 5-year survival (by Kaplan-Meier method) between predictor-defined high and low risk groups; and 2) a permutation test to evaluate the statistical significance of the cross-validated log-rank statistic. Here, we demonstrate strong stratification of 445 renal clear cell carcinoma tumors from The Cancer Genome Atlas (TCGA) into high and low risk groups using LASSO regression on RPPA data. Median cross-validated difference in 5-year overall survival was 32.8%, compared to 25.2% using a published receptor tyrosine kinase (RTK) prognostic signature (median hazard ratios of 3.3 and 2.4, respectively). Applicability and performance of our approach was demonstrated in three additional TCGA cohorts: ovarian serous cystadenocarcinoma (OVCA), sarcoma (SARC), and cutaneous melanoma (SKCM). The data-driven LASSO-based approach is versatile and well-suited for discovery of new protein/disease associations.

Identification of Implications of Angiogenesis and m6A Modification on Immunosuppression and Therapeutic Sensitivity in Low-Grade Glioma by Network Computational Analysis of Subtypes and Signatures

Angiogenesis is a complex process in the immunosuppressed low-grade gliomas (LGG) microenvironment and is regulated by multiple factors. N⁶-methyladenosine (m6A), modified by the m6A modification regulators (“writers” “readers” and “erasers”), can drive LGG formation. In the hypoxic environment of intracranial tumor immune microenvironment (TIME), m6A modifications in glioma stem cells are predominantly distributed around neovascularization and synergize with complex perivascular pathological ecology to mediate the immunosuppressive phenotype of TIME. The exact mechanism of this phenomenon remains unknown. Herein, we elucidated the relevance of the angiogenesis-related genes (ARGs) and m6A regulators (MAGs) and their influencing mechanism from a macro perspective. Based on the expression pattern of MAGs, we divided patients with LGG into two robust categories via consensus clustering, and further annotated the malignant related mechanisms and corresponding targeted agents. The two subgroups (CL1, CL2) demonstrated a significant correlation with prognosis and clinical-pathology features. Moreover, WGCNA has also uncovered the hub genes and related mechanisms of MAGs affecting clinical characters. Clustering analysis revealed a synergistic promoting effect of M6A and angiogenesis on immunosuppression. Based on the expression patterns of MAGs, we established a high-performance gene-signature (MASig). MASig revealed somatic mutational mechanisms by which MAGs affect the sensitivity to treatment in LGG patients. In conclusion, the MAGs were critical participants in the malignant process of LGG, with a vital potential in the prognosis stratification, prediction of outcome, and therapeutic sensitivity of LGG. Findings based on these strategies may facilitate the development of objective diagnosis and treatment systems to quantify patient survival and other outcomes, and in some cases, to identify potential unexplored targeted therapies.

XGBLC: an improved survival prediction model based on XGBoost

MOTIVATION

Survival analysis using gene expression profiles plays a crucial role in the interpretation of clinical research and assessment of disease therapy programs. Several prediction models have been developed to explore the relationship between patients' covariates and survival. However, the high-dimensional genomic features limit the prediction performance of the survival model. Thus, an accurate and reliable prediction model is necessary for survival analysis using high-dimensional genomic data.

RESULTS

In this study, we proposed an improved survival prediction model based on XGBoost framework called XGBLC, which used Lasso-Cox to enhance the ability to analyze high-dimensional genomic data. The novel first- and second-order gradient statistics of Lasso-Cox were defined to construct the loss function of XGBLC. We extensively tested our XGBLC algorithm on both simulated and real-world datasets, and estimated the performance of models with 5-fold cross-validation. Based on 20 cancer datasets from The Cancer Genome Atlas (TCGA), XGBLC outperforms five state-of-the-art survival methods in terms of C-index, Brier score and AUC. The results show that XGBLC still keeps good accuracy and robustness by comparing the performance on the simulated datasets with different scales. The developed prediction model would be beneficial for physicians to understand the effects of patient's genomic characteristics on survival and make personalized treatment decisions.

AVAILABILITY AND IMPLEMENTATION

The implementation of XGBLC algorithm based on R language is available at: https://github.com/lab319/XGBLC.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

An Effective Mental Stress State Detection and Evaluation System Using Minimum Number of Frontal Brain Electrodes

Currently, mental stress is a common social problem affecting people. Stress reduces human functionality during routine work and may lead to severe health defects. Detecting stress is important in education and industry to determine the efficiency of teaching, to improve education, and to reduce risks from human errors that might occur due to workers' stressful situations. Therefore, the early detection of mental stress using machine learning (ML) techniques is essential to prevent illness and health problems, improve quality of education, and improve industrial safety. The human brain is the main target of mental stress. For this reason, an ML system is proposed which investigates electroencephalogram (EEG) signal for thirty-six participants. Extracting useful features is essential for an efficient mental stress detection (MSD) system. Thus, this framework introduces a hybrid feature-set that feeds five ML classifiers to detect stress and non-stress states, and classify stress levels. To produce a reliable, practical, and efficient MSD system with a reduced number of electrodes, the proposed MSD scheme investigates the electrodes placements on different sites on the scalp and selects that site which has the higher impact on the accuracy of the system. Principal Component analysis is employed also, to reduce the features extracted from such electrodes to lower model complexity, where the optimal number of principal components is examined using sequential forward procedure. Furthermore, it examines the minimum number of electrodes placed on the site which has greater impact on stress detection and evaluation. To test the effectiveness of the proposed system, the results are compared with other feature extraction methods shown in literature. They are also compared with state-of-the-art techniques recorded for stress detection. The highest accuracies achieved in this study are 99.9%(sd = 0.015) and 99.26% (sd = 0.08) for identifying stress and non-stress states, and distinguishing between stress levels, respectively, using only two frontal brain electrodes for detecting stress and non-stress, and three frontal electrodes for evaluating stress levels respectively. The results show that the proposed system is reliable as the sensitivity is 99.9(0.064), 98.35(0.27), specificity is 99.94(0.02), 99.6(0.05), precision is 99.94(0.06), 98.9(0.23), and the diagnostics odd ratio (DOR) is ≥ 100 for detecting stress and non-stress, and evaluating stress levels respectively. This shows that the proposed framework has compelling performance and can be employed for stress detection and evaluation in medical, educational and industrial fields. Finally, the results verified the efficiency and reliability of the proposed system in predicting stress and non-stress on new patients, as the accuracy achieved 98.48% (sd = 1.12), sensitivity = 97.78% (sd = 1.84), specificity = 97.75% (sd = 2.05), precision = 99.26% (sd = 0.67), and DOR ≥ 100 using only two frontal electrodes.

Breast Cancer Diagnosis Using an Efficient CAD System Based on Multiple Classifiers

Breast cancer is one of the major health issues across the world. In this study, a new computer-aided detection (CAD) system is introduced. First, the mammogram images were enhanced to increase the contrast. Second, the pectoral muscle was eliminated and the breast was suppressed from the mammogram. Afterward, some statistical features were extracted. Next, k-nearest neighbor (k-NN) and decision trees classifiers were used to classify the normal and abnormal lesions. Moreover, multiple classifier systems (MCS) was constructed as it usually improves the classification results. The MCS has two structures, cascaded and parallel structures. Finally, two wrapper feature selection (FS) approaches were applied to identify those features, which influence classification accuracy. The two data sets (1) the mammographic image analysis society digital mammogram database (MIAS) and (2) the digital mammography dream challenge were combined together to test the CAD system proposed. The highest accuracy achieved with the proposed CAD system before FS was 99.7% using the Adaboosting of the J48 decision tree classifiers. The highest accuracy after FS was 100%, which was achieved with k-NN classifier. Moreover, the area under the curve (AUC) of the receiver operating characteristic (ROC) curve was equal to 1.0. The results showed that the proposed CAD system was able to accurately classify normal and abnormal lesions in mammogram samples.

Radiomics in Glioblastoma: Current Status and Challenges Facing Clinical Implementation

Radiomics analysis has had remarkable progress along with advances in medical imaging, most notability in central nervous system malignancies. Radiomics refers to the extraction of a large number of quantitative features that describe the intensity, texture and geometrical characteristics attributed to the tumor radiographic data. These features have been used to build predictive models for diagnosis, prognosis, and therapeutic response. Such models are being combined with clinical, biological, genetics and proteomic features to enhance reproducibility. Broadly, the four steps necessary for radiomic analysis are: (1) image acquisition, (2) segmentation or labeling, (3) feature extraction, and (4) statistical analysis. Major methodological challenges remain prior to clinical implementation. Essential steps include: adoption of an optimized standard imaging process, establishing a common criterion for performing segmentation, fully automated extraction of radiomic features without redundancy, and robust statistical modeling validated in the prospective setting. This review walks through these steps in detail, as it pertains to high grade gliomas. The impact on precision medicine will be discussed, as well as the challenges facing clinical implementation of radiomic in the current management of glioblastoma.

Prognostic Gene Discovery in Glioblastoma Patients using Deep Learning

This study aims to discover genes with prognostic potential for glioblastoma (GBM) patients’ survival in a patient group that has gone through standard of care treatments including surgeries and chemotherapies, using tumor gene expression at initial diagnosis before treatment. The Cancer Genome Atlas (TCGA) GBM gene expression data are used as inputs to build a deep multilayer perceptron network to predict patient survival risk using partial likelihood as loss function. Genes that are important to the model are identified by the input permutation method. Univariate and multivariate Cox survival models are used to assess the predictive value of deep learned features in addition to clinical, mutation, and methylation factors. The prediction performance of the deep learning method was compared to other machine learning methods including the ridge, adaptive Lasso, and elastic net Cox regression models. Twenty-seven deep-learned features are extracted through deep learning to predict overall survival. The top 10 ranked genes with the highest impact on these features are related to glioblastoma stem cells, stem cell niche environment, and treatment resistance mechanisms, including POSTN, TNR, BCAN, GAD1, TMSB15B, SCG3, PLA2G2A, NNMT, CHI3L1 and ELAVL4.

Information-Based Medicine in Glioma Patients: A Clinical Perspective

Glioma constitutes the most common type of primary brain tumor with a dismal survival, often measured in terms of months or years. The thin line between treatment effectiveness and patient harm underpins the importance of tailoring clinical management to the individual patient. Randomized trials have laid the foundation for many neuro-oncological guidelines. Despite this, their findings focus on group-level estimates. Given our current tools, we are limited in our ability to guide patients on what therapy is best for them as individuals, or even how long they should expect to survive. Machine learning, however, promises to provide the analytical support for personalizing treatment decisions, and deep learning allows clinicians to unlock insight from the vast amount of unstructured data that is collected on glioma patients. Although these novel techniques have achieved astonishing results across a variety of clinical applications, significant hurdles remain associated with the implementation of them in clinical practice. Future challenges include the assembly of well-curated cross-institutional datasets, improvement of the interpretability of machine learning models, and balancing novel evidence-based decision-making with the associated liability of automated inference. Although artificial intelligence already exceeds clinical expertise in a variety of applications, clinicians remain responsible for interpreting the implications of, and acting upon, each prediction.

Using multiple classifiers for predicting the risk of endovascular aortic aneurysm repair re-intervention through hybrid feature selection

Feature selection is essential in medical area; however, its process becomes complicated with the presence of censoring which is the unique character of survival analysis. Most survival feature selection methods are based on Cox's proportional hazard model, though machine learning classifiers are preferred. They are less employed in survival analysis due to censoring which prevents them from directly being used to survival data. Among the few work that employed machine learning classifiers, partial logistic artificial neural network with auto-relevance determination is a well-known method that deals with censoring and perform feature selection for survival data. However, it depends on data replication to handle censoring which leads to unbalanced and biased prediction results especially in highly censored data. Other methods cannot deal with high censoring. Therefore, in this article, a new hybrid feature selection method is proposed which presents a solution to high level censoring. It combines support vector machine, neural network, and K-nearest neighbor classifiers using simple majority voting and a new weighted majority voting method based on survival metric to construct a multiple classifier system. The new hybrid feature selection process uses multiple classifier system as a wrapper method and merges it with iterated feature ranking filter method to further reduce features. Two endovascular aortic repair datasets containing 91% censored patients collected from two centers were used to construct a multicenter study to evaluate the performance of the proposed approach. The results showed the proposed technique outperformed individual classifiers and variable selection methods based on Cox's model such as Akaike and Bayesian information criterions and least absolute shrinkage and selector operator in p values of the log-rank test, sensitivity, and concordance index. This indicates that the proposed classifier is more powerful in correctly predicting the risk of re-intervention enabling doctor in selecting patients' future follow-up plan.

Learning approaches to improve prediction of drug sensitivity in breast cancer patients

Predicting drug response to cancer disease is an important problem in modern clinical oncology that attracted increasing recent attention from various domains such as computational biology, machine learning, and data mining. Cancer patients respond differently to each cancer therapy owing to disease diversity, genetic factors, and environmental causes. Thus, oncologists aim to identify the effective therapies for cancer patients and avoid adverse drug reactions in patients. By predicting the drug response to cancer, oncologists gain full understanding of the effective treatments on each patient, which leads to better personalized treatment. In this paper, we present three learning approaches to improve the prediction of breast cancer patients' response to chemotherapy drug: the instance selection approach, the oversampling approach, and the hybrid approach. We evaluate the performance of our approaches and compare them against the baseline approach using the Area Under the ROC Curve (AUC) on clinical trial data, in addition to testing the stability of the approaches. Our experimental results show the stability of our approaches giving the highest AUC with statistical significance.

Aberrations in the Iron Regulatory Gene Signature Are Associated with Decreased Survival in Diffuse Infiltrating Gliomas

Iron is a tightly regulated micronutrient with no physiologic means of elimination and is necessary for cell division in normal tissue. Recent evidence suggests that dysregulation of iron regulatory proteins may play a role in cancer pathophysiology. We use public data from The Cancer Genome Atlas (TCGA) to study the association between survival and expression levels of 61 genes coding for iron regulatory proteins in patients with World Health Organization Grade II-III gliomas. Using a feature selection algorithm we identified a novel, optimized subset of eight iron regulatory genes (STEAP3, HFE, TMPRSS6, SFXN1, TFRC, UROS, SLC11A2, and STEAP4) whose differential expression defines two phenotypic groups with median survival differences of 52.3 months for patients with grade II gliomas (25.9 vs. 78.2 months, p< 10⁻³), 43.5 months for patients with grade III gliomas (43.9 vs. 87.4 months, p = 0.025), and 54.0 months when considering both grade II and III gliomas (79.9 vs. 25.9 months, p < 10⁻⁵).

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.