Multi-class cancer subtype classification based on gene expression signatures with reliability analysis

Pilot-Study to Explore Metabolic Signature of Type 2 Diabetes: A Pipeline of Tree-Based Machine Learning and Bioinformatics Techniques for Biomarkers Discovery

BACKGROUND

This study aims to identify unique metabolomics biomarkers associated with Type 2 Diabetes (T2D) and develop an accurate diagnostics model using tree-based machine learning (ML) algorithms integrated with bioinformatics techniques.

METHODS

Univariate and multivariate analyses such as fold change, a receiver operating characteristic curve (ROC), and Partial Least-Squares Discriminant Analysis (PLS-DA) were used to identify biomarker metabolites that showed significant concentration in T2D patients. Three tree-based algorithms [eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Adaptive Boosting (AdaBoost)] that demonstrated robustness in high-dimensional data analysis were used to create a diagnostic model for T2D.

RESULTS

As a result of the biomarker discovery process validated with three different approaches, Pyruvate, D-Rhamnose, AMP, pipecolate, Tetradecenoic acid, Tetradecanoic acid, Dodecanediothioic acid, Prostaglandin E3/D3 (isobars), ADP and Hexadecenoic acid were determined as potential biomarkers for T2D. Our results showed that the XGBoost model [accuracy = 0.831, F1-score = 0.845, sensitivity = 0.882, specificity = 0.774, positive predictive value (PPV) = 0.811, negative-PV (NPV) = 0.857 and Area under the ROC curve (AUC) = 0.887] had the slight highest performance measures.

CONCLUSIONS

ML integrated with bioinformatics techniques offers accurate and positive T2D candidate biomarker discovery. The XGBoost model can successfully distinguish T2D based on metabolites.

Cliques for the identification of gene signatures for colorectal cancer across population

Background

Colorectal cancer (CRC) is one of the most commonly diagnosed cancers worldwide. Studies have correlated risk of CRC development with dietary habits and environmental conditions. Gene signatures for any disease can identify the key biological processes, which is especially useful in studying cancer development. Such processes can be used to evaluate potential drug targets. Though recognition of CRC gene-signatures across populations is crucial to better understanding potential novel treatment options for CRC, it remains a challenging task.

Results

We developed a topological and biological feature-based network approach for identifying the gene signatures across populations. In this work, we propose a novel approach of using cliques to understand the variability within population. Cliques are more conserved and co-expressed, therefore allowing identification and comparison of cliques across a population which can help researchers study gene variations. Our study was based on four publicly available expression datasets belonging to four different populations across the world. We identified cliques of various sizes (0 to 7) across the four population networks. Cliques of size seven were further analyzed across populations for their commonality and uniqueness. Forty-nine common cliques of size seven were identified. These cliques were further analyzed based on their connectivity profiles. We found associations between the cliques and their connectivity profiles across networks. With these clique connectivity profiles (CCPs), we were able to identify the divergence among the populations, important biological processes (cell cycle, signal transduction, and cell differentiation), and related gene pathways. Therefore the genes identified in these cliques and their connectivity profiles can be defined as the gene-signatures across populations. In this work we demonstrate the power and effectiveness of cliques to study CRC across populations.

Conclusions

We developed a new approach where cliques and their connectivity profiles helped elucidate the variation and similarity in CRC gene profiles across four populations with unique dietary habits.

Lung cancer gene expression database analysis incorporating prior knowledge with support vector machine-based classification method

Background

A reliable and precise classification is essential for successful diagnosis and treatment of cancer. Gene expression microarrays have provided the high-throughput platform to discover genomic biomarkers for cancer diagnosis and prognosis. Rational use of the available bioinformation can not only effectively remove or suppress noise in gene chips, but also avoid one-sided results of separate experiment. However, only some studies have been aware of the importance of prior information in cancer classification.

Methods

Together with the application of support vector machine as the discriminant approach, we proposed one modified method that incorporated prior knowledge into cancer classification based on gene expression data to improve accuracy. A public well-known dataset, Malignant pleural mesothelioma and lung adenocarcinoma gene expression database, was used in this study. Prior knowledge is viewed here as a means of directing the classifier using known lung adenocarcinoma related genes. The procedures were performed by software R 2.80.

Results

The modified method performed better after incorporating prior knowledge. Accuracy of the modified method improved from 98.86% to 100% in training set and from 98.51% to 99.06% in test set. The standard deviations of the modified method decreased from 0.26% to 0 in training set and from 3.04% to 2.10% in test set.

Conclusion

The method that incorporates prior knowledge into discriminant analysis could effectively improve the capacity and reduce the impact of noise. This idea may have good future not only in practice but also in methodology.

Multiclass cancer classification by support vector machines with class-wise optimized genes and probability estimates

We investigate the multiclass classification of cancer microarray samples. In contrast to classification of two cancer types from gene expression data, multiclass classification of more than two cancer types are relatively hard and less studied problem. We used class-wise optimized genes with corresponding one-versus-all support vector machine (OVA-SVM) classifier to maximize the utilization of selected genes. Final prediction was made by using probability scores from all classifiers. We used three different methods of estimating probability from decision value. Among the three probability methods, Platt's approach was more consistent, whereas, isotonic approach performed better for datasets with unequal proportion of samples in different classes. Probability based decision does not only gives true and fair comparison between different one-versus-all (OVA) classifiers but also gives the possibility of using them for any post analysis. Several ensemble experiments, an example of post analysis, of the three probability methods were implemented to study their effect in improving the classification accuracy. We observe that ensemble did help in improving the predictive accuracy of cancer data sets especially involving unbalanced samples. Four-fold external stratified cross-validation experiment was performed on the six multiclass cancer datasets to obtain unbiased estimates of prediction accuracies. Analysis of class-wise frequently selected genes on two cancer datasets demonstrated that the approach was able to select important and relevant genes consistent to literature. This study demonstrates successful implementation of the framework of class-wise feature selection and multiclass classification for prediction of cancer subtypes on six datasets.

Biomarker panel diagnosis of thyroid cancer: a critical review

The accurate preoperative diagnosis of thyroid cancer continues to be a significant challenge for those individuals who present with nodular thyroid disease, particularly for tumors with indeterminate cytomorphological features by fine-needle aspiration biopsy. In an effort to develop improved diagnostic tools, a number of studies have investigated the discriminatory potential of many different RNA and protein molecules. However, no individual thyroid cancer biomarker has been found with sufficient sensitivity and specificity. Therefore, research focus has shifted to panels of multiple markers with the hope of improved performance and robustness. A panel comprised of GAL3, CK19 and HBME1 is by far the most studied to date and offers some improvement over individual marker performance alone. However, relatively few marker panels have been studied and their performances and application as diagnostic tests have not been consistently reported. We present a comprehensive review of molecular marker panel studies for thyroid tumors and current issues and challenges. In the future, studies evaluating larger numbers of biomarkers in large patient cohorts are required for the development and validation of a clinically applicable test.

Mistakes in validating the accuracy of a prediction classifier in high-dimensional but small-sample microarray data

A major interest in gene expression microarray studies is to develop an accurate classifier which can be adopted in clinical practice. The usage of large numbers of genes with small data samples may lead to overfitting in classification, and generate promising, but often nonreproducible results. Therefore, assessing the reproducibility of a classifier is necessary. Appropriate methods for validating a developed classifier and estimating its predicting accuracy are discussed. In addition, some mistakes that can arise in the cross validation process are reviewed using published articles in prominent medical journals, to prevent the indefinite results of a classifier development from leading to inappropriate treatment.

Microarrays for Cancer Diagnosis and Classification

Microarray analysis has yet to be widely accepted for diagnosis and classification of human cancers, despite the exponential increase in microarray studies reported in the literature. Among several methods available, a few refined approaches have evolved for the analysis of microarray data for cancer diagnosis. These include class comparison, class prediction and class discovery. Using as examples some of the major experimental contributions recently provided in the field of both hematological and solid tumors, we discuss the steps required to utilize microarray data to obtain general and reliable gene profiles that could be universally used in clinical laboratories. As we show, microarray technology is not only a new tool for the clinical lab but it can also improve the accuracy of the classical diagnostic techniques by suggesting novel tumor-specific markers. We then highlight the importance of publicly available microarray data and the development of their integrated analysis that may fulfill the promise that this new technology holds for cancer diagnosis and classification.

VPMCD: Variable interaction modeling approach for class discrimination in biological systems

Data classification algorithms applied for class prediction in computational biology literature are data specific and have shown varying degrees of performance. Different classes cannot be distinguished solely based on interclass distances or decision boundaries. We propose that inter-relations among the features be exploited for separating observations into specific classes. A new variable predictive model based class discrimination (VPMCD) method is described here. Three well established and proven data sets of varying statistical and biological significance are utilized as benchmark. The performance of the new method is compared with advanced classification algorithms. The new method performs better during different tests and shows higher stability and robustness. The VPMCD is observed to be a potentially strong classification approach and can be effectively extended to other data mining applications involving biological systems.

Translating microarray data for diagnostic testing in childhood leukaemia

Background

Recent findings from microarray studies have raised the prospect of a standardized diagnostic gene expression platform to enhance accurate diagnosis and risk stratification in paediatric acute lymphoblastic leukaemia (ALL). However, the robustness as well as the format for such a diagnostic test remains to be determined. As a step towards clinical application of these findings, we have systematically analyzed a published ALL microarray data set using Robust Multi-array Analysis (RMA) and Random Forest (RF).

Methods

We examined published microarray data from 104 ALL patients specimens, that represent six different subgroups defined by cytogenetic features and immunophenotypes. Using the decision-tree based supervised learning algorithm Random Forest (RF), we determined a small set of genes for optimal subgroup distinction and subsequently validated their predictive power in an independent patient cohort.

Results

We achieved very high overall ALL subgroup prediction accuracies of about 98%, and were able to verify the robustness of these genes in an independent panel of 68 specimens obtained from a different institution and processed in a different laboratory. Our study established that the selection of discriminating genes is strongly dependent on the analysis method. This may have profound implications for clinical use, particularly when the classifier is reduced to a small set of genes. We have demonstrated that as few as 26 genes yield accurate class prediction and importantly, almost 70% of these genes have not been previously identified as essential for class distinction of the six ALL subgroups.

Conclusion

Our finding supports the feasibility of qRT-PCR technology for standardized diagnostic testing in paediatric ALL and should, in conjunction with conventional cytogenetics lead to a more accurate classification of the disease. In addition, we have demonstrated that microarray findings from one study can be confirmed in an independent study, using an entirely independent patient cohort and with microarray experiments being performed by a different research team.

Artificial neural networks for diagnosis and survival prediction in colon cancer

ANNs are nonlinear regression computational devices that have been used for over 45 years in classification and survival prediction in several biomedical systems, including colon cancer. Described in this article is the theory behind the three-layer free forward artificial neural networks with backpropagation error, which is widely used in biomedical fields, and a methodological approach to its application for cancer research, as exemplified by colon cancer. Review of the literature shows that applications of these networks have improved the accuracy of colon cancer classification and survival prediction when compared to other statistical or clinicopathological methods. Accuracy, however, must be exercised when designing, using and publishing biomedical results employing machine-learning devices such as ANNs in worldwide literature in order to enhance confidence in the quality and reliability of reported data.

Modeling of SEB-induced host gene expression to correlate in vitro to in vivo responses

Detection of exposure to biological threat agents has relied on ever more sensitive methods for pathogen identification, but that usually requires pathogen proliferation to dangerous, near untreatable levels. Recent events have demonstrated that assessing exposure to a biological threat agent well in advance of onset of illness or at various stages post-exposure is invaluable among the diagnostic options. There is an urgent need for better diagnostic tools that will be sensitive, rapid, and unambiguous. Since human clinical cases of illness induced by biothreat agents are, fortunately, rare, use of animal models that closely mimic the human illness is the only in vivo option. Such studies can be very difficult and expensive; therefore, maximizing the information obtained from in vitro exposures to peripheral blood mononuclear cells (PBMCs) provide an opportunity to investigate dose/time variability in host responses. In our quest to study staphylococcal enterotoxin B (SEB) induced host gene expression patterns, we addressed two core issues using microarray analysis and predictive modeling. Our first objective was to determine gene expression patterns in human PBMCs exposed to SEB in vitro. Second, we compared the in vitro data with host responses gene expression patterns in vivo using PBMCs from an animal model of SEB intoxication that closely replicates the progression of illness in humans. We used cDNA microarrays to study global gene expression patterns in piglets intoxicated with SEB. We applied a supervised learning method for class prediction based on the k-nearest neighbor algorithm for the data obtained in piglets exposed to SEB in vivo against a training data set. This data set included gene expression profiles derived from in vitro exposures to eight different pathogens (Bacillus anthracis, Yersinia pestis, Brucella melitensis, SEB, cholera toxin, Clostridium botulinum toxin A, Venezuelan equine encephalitis, and Dengue-2) in PBMCs. We found that despite differences in gene expression profiles between in vitro and in vivo systems, there exists a subset of genes that show correlations between in vitro and in vivo exposures, which can be used as a predictor of exposure to SEB in vivo.

Evaluation of gene importance in microarray data based upon probability of selection

Background

Microarray devices permit a genome-scale evaluation of gene function. This technology has catalyzed biomedical research and development in recent years. As many important diseases can be traced down to the gene level, a long-standing research problem is to identify specific gene expression patterns linking to metabolic characteristics that contribute to disease development and progression. The microarray approach offers an expedited solution to this problem. However, it has posed a challenging issue to recognize disease-related genes expression patterns embedded in the microarray data. In selecting a small set of biologically significant genes for classifier design, the nature of high data dimensionality inherent in this problem creates substantial amount of uncertainty.

Results

Here we present a model for probability analysis of selected genes in order to determine their importance. Our contribution is that we show how to derive the P value of each selected gene in multiple gene selection trials based on different combinations of data samples and how to conduct a reliability analysis accordingly. The importance of a gene is indicated by its associated P value in that a smaller value implies higher information content from information theory. On the microarray data concerning the subtype classification of small round blue cell tumors, we demonstrate that the method is capable of finding the smallest set of genes (19 genes) with optimal classification performance, compared with results reported in the literature.

Conclusion

In classifier design based on microarray data, the probability value derived from gene selection based on multiple combinations of data samples enables an effective mechanism for reducing the tendency of fitting local data particularities.

Multi-class cancer classification by total principal component regression (TPCR) using microarray gene expression data

DNA microarray technology provides a promising approach to the diagnosis and prognosis of tumors on a genome-wide scale by monitoring the expression levels of thousands of genes simultaneously. One problem arising from the use of microarray data is the difficulty to analyze the high-dimensional gene expression data, typically with thousands of variables (genes) and much fewer observations (samples), in which severe collinearity is often observed. This makes it difficult to apply directly the classical statistical methods to investigate microarray data. In this paper, total principal component regression (TPCR) was proposed to classify human tumors by extracting the latent variable structure underlying microarray data from the augmented subspace of both independent variables and dependent variables. One of the salient features of our method is that it takes into account not only the latent variable structure but also the errors in the microarray gene expression profiles (independent variables). The prediction performance of TPCR was evaluated by both leave-one-out and leave-half-out cross-validation using four well-known microarray datasets. The stabilities and reliabilities of the classification models were further assessed by re-randomization and permutation studies. A fast kernel algorithm was applied to decrease the computation time dramatically. (MATLAB source code is available upon request.).

SED, a normalization free method for DNA microarray data analysis

BACKGROUND

Analysis of DNA microarray data usually begins with a normalization step where intensities of different arrays are adjusted to the same scale so that the intensity levels from different arrays can be compared with one other. Both simple total array intensity-based as well as more complex "local intensity level" dependent normalization methods have been developed, some of which are widely used. Much less developed methods for microarray data analysis include those that bypass the normalization step and therefore yield results that are not confounded by potential normalization errors.

RESULTS

Instead of focusing on the raw intensity levels, we developed a new method for microarray data analysis that maps each gene's expression intensity level to a high dimensional space of SEDs (Signs of Expression Difference), the signs of the expression intensity difference between a given gene and every other gene on the array. Since SED are unchanged under any monotonic transformation of intensity levels, the SED based method is normalization free. When tested on a multi-class tumor classification problem, simple Naive Bayes and Nearest Neighbor methods using the SED approach gave results comparable with normalized intensity-based algorithms. Furthermore, a high percentage of classifiers based on a single gene's SED gave good classification results, suggesting that SED does capture essential information from the intensity levels.

CONCLUSION

The results of testing this new method on multi-class tumor classification problems suggests that the SED-based, normalization-free method of microarray data analysis is feasible and promising.