Recursive partitioning for tumor classification with gene expression microarray data

A review of bioinformatics tools and web servers in different microarray platforms used in cancer research

Over the past decade, conventional lab work strategies have gradually shifted from being limited to a laboratory setting towards a bioinformatics era to help manage and process the vast amounts of data generated by omics technologies. The present work outlines the latest contributions of bioinformatics in analyzing microarray data and their application to cancer. We dissect different microarray platforms and their use in gene expression in cancer models. We highlight how computational advances empowered the microarray technology in gene expression analysis. The study on protein-protein interaction databases classified into primary, derived, meta-database, and prediction databases describes the strategies to curate and predict novel interaction networks in silico. In addition, we summarize the areas of bioinformatics where neural graph networks are currently being used, such as protein functions, protein interaction prediction, and in silico drug discovery and development. We also discuss the role of deep learning as a potential tool in the prognosis, diagnosis, and treatment of cancer. Integrating these resources efficiently, practically, and ethically is likely to be the most challenging task for the healthcare industry over the next decade; however, we believe that it is achievable in the long term.

Mechanism-Centric Approaches for Biomarker Detection and Precision Therapeutics in Cancer

Biomarker discovery is at the heart of personalized treatment planning and cancer precision therapeutics, encompassing disease classification and prognosis, prediction of treatment response, and therapeutic targeting. However, many biomarkers represent passenger rather than driver alterations, limiting their utilization as functional units for therapeutic targeting. We suggest that identification of driver biomarkers through mechanism-centric approaches, which take into account upstream and downstream regulatory mechanisms, is fundamental to the discovery of functionally meaningful markers. Here, we examine computational approaches that identify mechanism-centric biomarkers elucidated from gene co-expression networks, regulatory networks (e.g., transcriptional regulation), protein-protein interaction (PPI) networks, and molecular pathways. We discuss their objectives, advantages over gene-centric approaches, and known limitations. Future directions highlight the importance of input and model interpretability, method and data integration, and the role of recently introduced technological advantages, such as single-cell sequencing, which are central for effective biomarker discovery and time-cautious precision therapeutics.

A Holistic Performance Comparison for Lung Cancer Classification Using Swarm Intelligence Techniques

In the field of bioinformatics, feature selection in classification of cancer is a primary area of research and utilized to select the most informative genes from thousands of genes in the microarray. Microarray data is generally noisy, is highly redundant, and has an extremely asymmetric dimensionality, as the majority of the genes present here are believed to be uninformative. The paper adopts a methodology of classification of high dimensional lung cancer microarray data utilizing feature selection and optimization techniques. The methodology is divided into two stages; firstly, the ranking of each gene is done based on the standard gene selection techniques like Information Gain, Relief–F test, Chi-square statistic, and T-statistic test. As a result, the gathering of top scored genes is assimilated, and a new feature subset is obtained. In the second stage, the new feature subset is further optimized by using swarm intelligence techniques like Grasshopper Optimization (GO), Moth Flame Optimization (MFO), Bacterial Foraging Optimization (BFO), Krill Herd Optimization (KHO), and Artificial Fish Swarm Optimization (AFSO), and finally, an optimized subset is utilized. The selected genes are used for classification, and the classifiers used here are Naïve Bayesian Classifier (NBC), Decision Trees (DT), Support Vector Machines (SVM), and K-Nearest Neighbour (KNN). The best results are shown when Relief-F test is computed with AFSO and classified with Decision Trees classifier for hundred genes, and the highest classification accuracy of 99.10% is obtained.

Colon cancer data analysis by chameleon algorithm

Detecting the key differential genes of colon cancers is very important to tell colon cancer patients from normal people. A gene selection algorithm for colon cancers is proposed by using the dynamic modeling properties of chameleon algorithm and its capability to discover any arbitrary shape clusters. This chameleon algorithm based gene selection algorithm comprises three steps. The first step is to select those genes with higher Fisher function values as candidate genes. The second step is to detect gene groups by using chameleon algorithm based on Euclidean distance. The third step is to select the most important gene from each gene cluster to comprise the gene subset by using the information index to classification of each gene. After that the chameleon algorithm is used to detect groups of colon cancer patients and normal people only with genes in gene subset. The final clustering accuracy of chameleon algorithm with the selected genes is up to 85.48%. The clustering analysis to colon cancer data and the comparisons to the other related studies demonstrate that the proposed algorithm is effective in detecting the differential genes of colon cancers.

Using recursive partitioning approach to select tumor-associated antigens in immunodiagnosis of gastric adenocarcinoma

The present study aimed to select anti‐tumor‐associated antigen (TAA) autoantibodies as biomarkers in the immunodiagnosis of gastric adenocarcinoma (GAC) by the recursive partitioning approach (RPA) and further construct and evaluate a predictive model. A case‐control study was designed including 407 GAC patients as the case group and 407 normal controls. In addition, 67 serial serum samples from 25 GAC patients were collected at different time points before and after gastrectomy treatment. Autoantibodies against 14 TAA were measured in sera from all subjects by enzyme immunoassay. Finally, RPA resulted in the selection of nine‐panel TAA (c‐Myc, p16, HSPD1, PTEN, p53, NPM1, ENO1, p62, HCC1.4) from all detected TAA in the case‐control study; the classification tree based on this nine‐TAA panel had area under curve (AUC) of 0.857, sensitivity of 71.5% and specificity of 71.3%; The optimal panel also can identify GAC patients at an early stage from normal individuals, with AUC of 0.737, sensitivity of 64.9% and specificity of 70.5%. However, frequencies of the nine autoantibodies showed no correlation with GAC stage, tumor size, lymphatic metastasis or differentiation. GAC patients positive for more than two autoantibodies in the nine‐TAA panel had a worse prognosis than that of the GAC patients positive for no or one antibody. Titers of 10 autoantibodies in serial serum samples were significantly higher in GAC patients after surgical resection than before. In conclusion, this study showed that the panel of nine multiple TAAs could enhance the detection of anti‐TAA antibodies in GAC, and may be potential prognostic biomarkers in GAC.

Optimizing community-level surveillance data for pediatric asthma management

Community-level approaches for pediatric asthma management rely on locally collected information derived primarily from two sources: claims records and school-based surveys. We combined claims and school-based surveillance data, and examined the asthma-related risk patterns among adolescent students. Symptom data collected from school-based asthma surveys conducted in Oakland, CA were used for case identification and determination of severity levels for students (high and low). Survey data were matched to Medicaid claims data for all asthma-related health care encounters for the year prior to the survey. We then employed recursive partitioning to develop classification trees that identified patterns of demographics and healthcare utilization associated with severity. A total of 561 students had complete matched data; 86.1% were classified as high-severity, and 13.9% as low-severity asthma. The classification tree consisted of eight subsets: three indicating high severity and five indicating low severity. The risk subsets highlighted varying combinations of non-specific demographic and socioeconomic predictors of asthma prevalence, morbidity and severity. For example, the subset with the highest class-prior probability (92.1%) predicted high-severity asthma and consisted of students without prescribed rescue medication, but with at least one in-clinic nebulizer treatment. The predictive accuracy of the tree-based model was approximately 66.7%, with an estimated 91.1% of high-severity cases and 42.3% of low-severity cases correctly predicted. Our analysis draws on the strengths of two complementary datasets to provide community-level information on children with asthma, and demonstrates the utility of recursive partitioning methods to explore a combination of features that convey asthma severity.

Partitioning of functional gene expression data using principal points

Background

DNA microarrays offer motivation and hope for the simultaneous study of variations in multiple genes. Gene expression is a temporal process that allows variations in expression levels with a characterized gene function over a period of time. Temporal gene expression curves can be treated as functional data since they are considered as independent realizations of a stochastic process. This process requires appropriate models to identify patterns of gene functions. The partitioning of the functional data can find homogeneous subgroups of entities for the massive genes within the inherent biological networks. Therefor it can be a useful technique for the analysis of time-course gene expression data. We propose a new self-consistent partitioning method of functional coefficients for individual expression profiles based on the orthonormal basis system.

Results

A principal points based functional partitioning method is proposed for time-course gene expression data. The method explores the relationship between genes using Legendre coefficients as principal points to extract the features of gene functions. Our proposed method provides high connectivity in connectedness after clustering for simulated data and finds a significant subsets of genes with the increased connectivity. Our approach has comparative advantages that fewer coefficients are used from the functional data and self-consistency of principal points for partitioning. As real data applications, we are able to find partitioned genes through the gene expressions found in budding yeast data and Escherichia coli data.

Conclusions

The proposed method benefitted from the use of principal points, dimension reduction, and choice of orthogonal basis system as well as provides appropriately connected genes in the resulting subsets. We illustrate our method by applying with each set of cell-cycle-regulated time-course yeast genes and E. coli genes. The proposed method is able to identify highly connected genes and to explore the complex dynamics of biological systems in functional genomics.

Genotype-specific physiological and transcriptomic responses to drought stress in Setaria italica (an emerging model for Panicoideae grasses)

Understanding drought-tolerance mechanisms and identifying genetic dominance are important for crop improvement. Setaria italica, which is extremely drought-tolerant, has been regarded as a model plant for studying stress biology. Moreover, different genotypes of S. italica have evolved various drought-tolerance/avoidance mechanisms that should be elucidated. Physiological and transcriptomic comparisons between drought-tolerant S. italica cultivar 'Yugu1' and drought-sensitive 'An04' were conducted. 'An04' had higher yields and more efficient photosystem activities than 'Yugu1' under well-watered conditions, and this was accompanied by positive brassinosteroid regulatory actions. However, 'An04's growth advantage was severely repressed by drought, while 'Yugu1' maintained normal growth under a water deficiency. High-throughput sequencing suggested that the S. italica transcriptome was severely remodelled by genotype × environment interactions. Expression profiles of genes related to phytohormone metabolism and signalling, transcription factors, detoxification, and other stress-related proteins were characterised, revealing genotype-dependent and -independent drought responses in different S. italica genotypes. Combining our data with drought-tolerance-related QTLs, we identified 20 candidate genes that contributed to germination and early seedling' drought tolerance in S. italica. Our analysis provides a comprehensive picture of how different S. italica genotypes respond to drought, and may be used for the genetic improvement of drought tolerance in Poaceae crops.

Global Gene Expression Profiling: A Complement to Conventional Histopathologic Analysis of Neoplasia

Transcriptional profiling of entire tumors has yielded considerable insight into the molecular mechanisms of heterogeneous cell populations within different types of neoplasms. The data thus acquired can be further refined by microdissection methods that enable the analyses of subpopulations of neoplastic cells. Separation of the various components of a neoplasm (i.e., stromal cells, inflammatory infiltrates, and blood vessels) has been problematic, primarily because of a paucity of tools for accurate microdissection. The advent of laser capture microdissection combined with powerful tools of linear amplification of RNA and high-throughput microarray-based assays have allowed the transcriptional mapping of intricate and highly complex networks within pure populations of neoplastic cells. With this approach, specific “molecular signatures” can be assigned to tumors of distinct or even similar histomorphology, thereby aiding the desired objective of pattern recognition, tumor classification, and prognostication. This review highlights the potential benefits of global gene expression profiling of tumor cells as a complement to conventional histopathologic analyses.

A novel comparative pattern analysis approach identifies chronic alcohol mediated dysregulation of transcriptomic dynamics during liver regeneration

BACKGROUND

Liver regeneration is inhibited by chronic ethanol consumption and this impaired repair response may contribute to the risk for alcoholic liver disease. We developed and applied a novel data analysis approach to assess the effect of chronic ethanol intake in the mechanisms responsible for liver regeneration. We performed a time series transcriptomic profiling study of the regeneration response after 2/3rd partial hepatectomy (PHx) in ethanol-fed and isocaloric control rats.

RESULTS

We developed a novel data analysis approach focusing on comparative pattern counts (COMPACT) to exhaustively identify the dominant and subtle differential expression patterns. Approximately 6500 genes were differentially regulated in Ethanol or Control groups within 24 h after PHx. Adaptation to chronic ethanol intake significantly altered the immediate early gene expression patterns and nearly completely abrogated the cell cycle induction in hepatocytes post PHx. The patterns highlighted by COMPACT analysis contained several non-parenchymal cell specific markers indicating their aberrant transcriptional response as a novel mechanism through which chronic ethanol intake deregulates the integrated liver tissue response.

CONCLUSIONS

Our novel comparative pattern analysis revealed new insights into ethanol-mediated molecular changes in non-parenchymal liver cells as a possible contribution to the defective liver regeneration phenotype. The results revealed for the first time an ethanol-induced shift of hepatic stellate cells from a pro-regenerative phenotype to that of an anti-regenerative state after PHx. Our results can form the basis for novel interventions targeting the non-parenchymal cells in normalizing the dysfunctional repair response process in alcoholic liver disease. Our approach is illustrated online at http://compact.jefferson.edu .

The feature selection bias problem in relation to high-dimensional gene data

OBJECTIVE

Feature selection is a technique widely used in data mining. The aim is to select the best subset of features relevant to the problem being considered. In this paper, we consider feature selection for the classification of gene datasets. Gene data is usually composed of just a few dozen objects described by thousands of features. For this kind of data, it is easy to find a model that fits the learning data. However, it is not easy to find one that will simultaneously evaluate new data equally well as learning data. This overfitting issue is well known as regards classification and regression, but it also applies to feature selection.

METHODS AND MATERIALS

We address this problem and investigate its importance in an empirical study of four feature selection methods applied to seven high-dimensional gene datasets. We chose datasets that are well studied in the literature-colon cancer, leukemia and breast cancer. All the datasets are characterized by a significant number of features and the presence of exactly two decision classes. The feature selection methods used are ReliefF, minimum redundancy maximum relevance, support vector machine-recursive feature elimination and relaxed linear separability.

RESULTS

Our main result reveals the existence of positive feature selection bias in all 28 experiments (7 datasets and 4 feature selection methods). Bias was calculated as the difference between validation and test accuracies and ranges from 2.6% to as much as 41.67%. The validation accuracy (biased accuracy) was calculated on the same dataset on which the feature selection was performed. The test accuracy was calculated for data that was not used for feature selection (by so called external cross-validation).

CONCLUSIONS

This work provides evidence that using the same dataset for feature selection and learning is not appropriate. We recommend using cross-validation for feature selection in order to reduce selection bias.

Identifying ion channel genes related to cardiomyopathy using a novel decision forest strategy

Ion channels play many crucial functions in life. Their dysfunction may lead to a number of diseases, such as arrhythmia and beta cell dysfunction. In this study, we firstly selected the ion channel gene expression profiles using a dimensionality reduction method. After that, we applied a novel decision forest strategy to mine cardiomyopathy related ion channel genes. The novel proposed Zi integrated the information of the decision trees' height and the frequency at which a gene was located in the tree. It achieved a much higher ability of feature selection. In the result, 26 cardiomyopathy related ion channel genes were identified. Their Zi were higher than the threshold Z*. Furthermore, most of these genes had been reported to have relationships with cardiomyopathies. In conclusion, our proposed decision forest strategy had a better classification performance. Our result can provide a theoretical basis for cardiovascular researchers.

Identifying ultrasound and clinical features of breast cancer molecular subtypes by ensemble decision

Breast cancer is molecularly heterogeneous and categorized into four molecular subtypes: Luminal-A, Luminal-B, HER2-amplified and Triple-negative. In this study, we aimed to apply an ensemble decision approach to identify the ultrasound and clinical features related to the molecular subtypes. We collected ultrasound and clinical features from 1,000 breast cancer patients and performed immunohistochemistry on these samples. We used the ensemble decision approach to select unique features and to construct decision models. The decision model for Luminal-A subtype was constructed based on the presence of an echogenic halo and post-acoustic shadowing or indifference. The decision model for Luminal-B subtype was constructed based on the absence of an echogenic halo and vascularity. The decision model for HER2-amplified subtype was constructed based on the presence of post-acoustic enhancement, calcification, vascularity and advanced age. The model for Triple-negative subtype followed two rules. One was based on irregular shape, lobulate margin contour, the absence of calcification and hypovascularity, whereas the other was based on oval shape, hypovascularity and micro-lobulate margin contour. The accuracies of the models were 83.8%, 77.4%, 87.9% and 92.7%, respectively. We identified specific features of each molecular subtype and expanded the scope of ultrasound for making diagnoses using these decision models.

Urban park-related risks forGiardiaspp. infection in dogs

Giardiaspp. is a common gastrointestinal (GI) parasite of multiple host species, including dogs and humans, with the potential for zoonotic transmission. The risk of GI parasitism in dogs (includingGiardiaspp.) may increase with park use in urban areas. This study aimed to (1) determine whether park attendance is a risk factor forGiardiaspp. infection in dogs and (2) characterize the behavioural and demographic risk factors forGiardiaspp. infection in park-attending and non-park-attending dogs. From August to September 2012, a total of 1293 dog owners completed a survey and 860 corresponding dog faecal samples were collected. Dog faeces were screened forGiardiaspp. using a direct immunofluorescence assay and associations assessed among behaviours, demographics, andGiardiaspp. infection. Main results included off-leash and swimming frequencies within parks as significantly positively associated withGiardiaspp. infection in dogs. Dog-owner age was negatively associated with off-leash and swimming frequencies in parks. The results suggest some recreational behaviours in parks and certain demographics are risk factors for parasitism in pet dogs.

Fuzzy support vector machine: an efficient rule-based classification technique for microarrays

BACKGROUND

The abundance of gene expression microarray data has led to the development of machine learning algorithms applicable for tackling disease diagnosis, disease prognosis, and treatment selection problems. However, these algorithms often produce classifiers with weaknesses in terms of accuracy, robustness, and interpretability. This paper introduces fuzzy support vector machine which is a learning algorithm based on combination of fuzzy classifiers and kernel machines for microarray classification.

RESULTS

Experimental results on public leukemia, prostate, and colon cancer datasets show that fuzzy support vector machine applied in combination with filter or wrapper feature selection methods develops a robust model with higher accuracy than the conventional microarray classification models such as support vector machine, artificial neural network, decision trees, k nearest neighbors, and diagonal linear discriminant analysis. Furthermore, the interpretable rule-base inferred from fuzzy support vector machine helps extracting biological knowledge from microarray data.

CONCLUSIONS

Fuzzy support vector machine as a new classification model with high generalization power, robustness, and good interpretability seems to be a promising tool for gene expression microarray classification.

A classification framework applied to cancer gene expression profiles

Classification of cancer based on gene expression has provided insight into possible treatment strategies. Thus, developing machine learning methods that can successfully distinguish among cancer subtypes or normal versus cancer samples is important. This work discusses supervised learning techniques that have been employed to classify cancers. Furthermore, a two-step feature selection method based on an attribute estimation method (e.g., ReliefF) and a genetic algorithm was employed to find a set of genes that can best differentiate between cancer subtypes or normal versus cancer samples. The application of different classification methods (e.g., decision tree, k-nearest neighbor, support vector machine (SVM), bagging, and random forest) on 5 cancer datasets shows that no classification method universally outperforms all the others. However, k-nearest neighbor and linear SVM generally improve the classification performance over other classifiers. Finally, incorporating diverse types of genomic data (e.g., protein-protein interaction data and gene expression) increase the prediction accuracy as compared to using gene expression alone.

REACTIN: regulatory activity inference of transcription factors underlying human diseases with application to breast cancer

BACKGROUND

Genetic alterations of transcription factors (TFs) have been implicated in the tumorigenesis of cancers. In many cancers, alteration of TFs results in aberrant activity of them without changing their gene expression level. Gene expression data from microarray or RNA-seq experiments can capture the expression change of genes, however, it is still challenge to reveal the activity change of TFs.

RESULTS

Here we propose a method, called REACTIN (REgulatory ACTivity INference), which integrates TF binding data with gene expression data to identify TFs with significantly differential activity between disease and normal samples. REACTIN successfully detect differential activity of estrogen receptor (ER) between ER+ and ER- samples in 10 breast cancer datasets. When applied to compare tumor and normal breast samples, it reveals TFs that are critical for carcinogenesis of breast cancer. Moreover, Reaction can be utilized to identify transcriptional programs that are predictive to patient survival time of breast cancer patients.

CONCLUSIONS

REACTIN provides a useful tool to investigate regulatory programs underlying a biological process providing the related case and control gene expression data. Considering the enormous amount of cancer gene expression data and the increasingly accumulating ChIP-seq data, we expect wide application of REACTIN for revealing the regulatory mechanisms of various diseases.

Classification of tumor samples from expression data using decision trunks

We present a novel machine learning approach for the classification of cancer samples using expression data. We refer to the method as “decision trunks,” since it is loosely based on decision trees, but contains several modifications designed to achieve an algorithm that: (1) produces smaller and more easily interpretable classifiers than decision trees; (2) is more robust in varying application scenarios; and (3) achieves higher classification accuracy. The decision trunk algorithm has been implemented and tested on 26 classification tasks, covering a wide range of cancer forms, experimental methods, and classification scenarios. This comprehensive evaluation indicates that the proposed algorithm performs at least as well as the current state of the art algorithms in terms of accuracy, while producing classifiers that include on average only 2–3 markers. We suggest that the resulting decision trunks have clear advantages over other classifiers due to their transparency, interpretability, and their correspondence with human decision-making and clinical testing practices.

Differential combinatorial regulatory network analysis related to venous metastasis of hepatocellular carcinoma

Background

Hepatocellular carcinoma (HCC) is one of the most fatal cancers in the world, and metastasis is a significant cause to the high mortality in patients with HCC. However, the molecular mechanism behind HCC metastasis is not fully understood. Study of regulatory networks may help investigate HCC metastasis in the way of systems biology profiling.

Methods

By utilizing both sequence information and parallel microRNA(miRNA) and mRNA expression data on the same cohort of HBV related HCC patients without or with venous metastasis, we constructed combinatorial regulatory networks of non-metastatic and metastatic HCC which contain transcription factor(TF) regulation and miRNA regulation. Differential regulation patterns, classifying marker modules, and key regulatory miRNAs were analyzed by comparing non-metastatic and metastatic networks.

Results

Globally TFs accounted for the main part of regulation while miRNAs for the minor part of regulation. However miRNAs displayed a more active role in the metastatic network than in the non-metastatic one. Seventeen differential regulatory modules discriminative of the metastatic status were identified as cumulative-module classifier, which could also distinguish survival time. MiR-16, miR-30a, Let-7e and miR-204 were identified as key miRNA regulators contributed to HCC metastasis.

Conclusion

In this work we demonstrated an integrative approach to conduct differential combinatorial regulatory network analysis in the specific context venous metastasis of HBV-HCC. Our results proposed possible transcriptional regulatory patterns underlying the different metastatic subgroups of HCC. The workflow in this study can be applied in similar context of cancer research and could also be extended to other clinical topics.

Gaining Confidence on Molecular Classification through Consensus Modeling and Validation

Current advances in genomics, proteomics, and metabonomics would result in a constellation of benefits in human health. Classification applying supervised learning methods to omics data as one of the molecular classification approaches has enjoyed its growing role in clinical application. However, the utility of a molecular classifier will not be fully appreciated unless its quality is carefully validated. A clinical omics data is usually noisy with the number of independent variables far more than the number of subjects and, possibly, with a skewed subject distribution. Given that, the consensus approach holds an advantage over a single classifier. Thus, the focus of this review is mainly placed on how validating a molecular classifier using Decision Forest (DF), a robust consensus approach. We recommended that a molecular classifier has to be assessed with respect to overall prediction accuracy, prediction confidence and chance correlation, which can be readily achieved in DF. The commonalities and differences between external validation and cross-validation are also discussed for perspective use of these methods to validate a DF classifier. In addition, the advantages of using consensus approaches for identification of potential biomarkers are also rationalized. Although specific DF examples are used in this review, the provided rationales and recommendations should be equally applicable to other consensus methods.

Identifying representative trees from ensembles

Tree-based methods have become popular for analyzing complex data structures where the primary goal is risk stratification of patients. Ensemble techniques improve the accuracy in prediction and address the instability in a single tree by growing an ensemble of trees and aggregating. However, in the process, individual trees get lost. In this paper, we propose a methodology for identifying the most representative trees in an ensemble on the basis of several tree distance metrics. Although our focus is on binary outcomes, the methods are applicable to censored data as well. For any two trees, the distance metrics are chosen to (1) measure similarity of the covariates used to split the trees; (2) reflect similar clustering of patients in the terminal nodes of the trees; and (3) measure similarity in predictions from the two trees. Whereas the latter focuses on prediction, the first two metrics focus on the architectural similarity between two trees. The most representative trees in the ensemble are chosen on the basis of the average distance between a tree and all other trees in the ensemble. Out-of-bag estimate of error rate is obtained using neighborhoods of representative trees. Simulations and data examples show gains in predictive accuracy when averaging over such neighborhoods. We illustrate our methods using a dataset of kidney cancer treatment receipt (binary outcome) and a second dataset of breast cancer survival (censored outcome).

Biomarker classifiers for identifying susceptible subpopulations for treatment decisions

AIM

A main goal of pharmacogenomics is to develop genomic signatures to predict patients' responses to a drug or therapy for treatment decisions. Identification of patients who would have no beneficial effect or have the risk of developing adverse effects from an unnecessary treatment could save enormous cost in the healthcare system and clinical trials. This article presents an approach for developing a biomarker classifier for identifying a fraction of susceptible patients, who should be spared unnecessary treatment prior to treatment.

MATERIALS & METHODS

The identification of susceptible patients involves two steps. The first step is to identify biomarkers of susceptibility from a mixture of biomarkers of susceptibility and biomarkers of response; the second step is to develop a classifier using an ensemble classification algorithm, as the number of susceptible patients is generally much smaller than the number of nonsusceptible patients.

RESULTS

Selection of the biomarkers of susceptibility is essential to achieve good prediction accuracy. The ensemble algorithm significantly improves the prediction accuracy compared with the standard classifiers.

CONCLUSION

The study shows that classifiers developed based on the biomarkers obtained by comparing the genomic profiles of responders to those of nonresponders may lead to a high misclassification error rate. Classifiers to identify a small fraction of the subpopulation should take imbalanced class sizes into consideration. A large sample size may be needed in order to ensure detection of a sufficient number of biomarkers and a sufficient number of susceptible subjects for classifier development and validation.

INTERRELATED TWO-WAY CLUSTERING AND ITS APPLICATION ON GENE EXPRESSION DATA

Microarray technologies are capable of simultaneously measuring the signals for thousands of messenger RNAs and large numbers of proteins from single samples. Arrays are now widely used in basic biomedical research for mRNA expression profiling and are increasingly being used to explore patterns of gene expression in clinical research. Most research has focused on the interpretation of the meaning of the microarray data which are transformed into gene expression matrices where usually the rows represent genes, the columns represent various samples. Clustering samples can be done by analyzing and eliminating of irrelevant genes. However, majority methods are supervised (or assisted by domain knowledge), less attention has been paid on unsupervised approaches which are important when little domain knowledge is available. In this paper, we present a new framework for unsupervised analysis of gene expression data, which applies an interrelated two-way clustering approach on the gene expression matrices. The goal of clustering is to identify important genes and perform cluster discovery on samples. The advantage of this approach is that we can dynamically manipulate the relationship between the gene clusters and sample groups while conducting an iterative clustering through both of them. The performance of the proposed method with various gene expression data sets is also illustrated.

Classification of dendritic cell phenotypes from gene expression data

Background

The selection of relevant genes for sample classification is a common task in many gene expression studies. Although a number of tools have been developed to identify optimal gene expression signatures, they often generate gene lists that are too long to be exploited clinically. Consequently, researchers in the field try to identify the smallest set of genes that provide good sample classification. We investigated the genome-wide expression of the inflammatory phenotype in dendritic cells. Dendritic cells are a complex group of cells that play a critical role in vertebrate immunity. Therefore, the prediction of the inflammatory phenotype in these cells may help with the selection of immune-modulating compounds.

Results

A data mining protocol was applied to microarray data for murine cell lines treated with various inflammatory stimuli. The learning and validation data sets consisted of 155 and 49 samples, respectively. The data mining protocol reduced the number of probe sets from 5,802 to 10, then from 10 to 6 and finally from 6 to 3. The performances of a set of supervised classification models were compared. The best accuracy, when using the six following genes --Il12b, Cd40, Socs3, Irgm1, Plin2 and Lgals3bp-- was obtained by Tree Augmented Naïve Bayes and Nearest Neighbour (91.8%). Using the smallest set of three genes --Il12b, Cd40 and Socs3-- the performance remained satisfactory and the best accuracy was with Support Vector Machine (95.9%). These data mining models, using data for the genes Il12b, Cd40 and Socs3, were validated with a human data set consisting of 27 samples. Support Vector Machines (71.4%) and Nearest Neighbour (92.6%) gave the worst performances, but the remaining models correctly classified all the 27 samples.

Conclusions

The genes selected by the data mining protocol proposed were shown to be informative for discriminating between inflammatory and steady-state phenotypes in dendritic cells. The robustness of the data mining protocol was confirmed by the accuracy for a human data set, when using only the following three genes: Il12b, Cd40 and Socs3. In summary, we analysed the longitudinal pattern of expression in dendritic cells stimulated with activating agents with the aim of identifying signatures that would predict or explain the dentritic cell response to an inflammatory agent.

A cDNA microarray gene expression data classifier for clinical diagnostics based on graph theory

Despite great advances in discovering cancer molecular profiles, the proper application of microarray technology to routine clinical diagnostics is still a challenge. Current practices in the classification of microarrays' data show two main limitations: the reliability of the training data sets used to build the classifiers, and the classifiers' performances, especially when the sample to be classified does not belong to any of the available classes. In this case, state-of-the-art algorithms usually produce a high rate of false positives that, in real diagnostic applications, are unacceptable. To address this problem, this paper presents a new cDNA microarray data classification algorithm based on graph theory and is able to overcome most of the limitations of known classification methodologies. The classifier works by analyzing gene expression data organized in an innovative data structure based on graphs, where vertices correspond to genes and edges to gene expression relationships. To demonstrate the novelty of the proposed approach, the authors present an experimental performance comparison between the proposed classifier and several state-of-the-art classification algorithms.

Systematic investigation of global coordination among mRNA and protein in cellular society

Background

Cell functions depend on molecules organized in the cellular society. Two basic components are mRNA molecules and proteins. The interactions within and between those two components are crucial for carrying out sophisticated cell functions. The interplay can be analyzed by comparing expression levels of mRNA and proteins. This is critical for understanding the molecular interactions, (post-) transcriptional regulations and conservation of co-expression between mRNAs and proteins. By using high-throughput transcriptome and proteome data, this study aims to systematically investigate the general picture of such expression correlations. We analyze four groups of correlations: (i) transcript levels of different genes, (ii) protein levels of different genes, (iii) mRNA levels with protein levels of different genes and (iv) mRNA levels with protein levels of same genes. This helps to obtain global insights into the stability and variability of co-expression and correlation of mRNA and protein levels.

Results

Analysis of the simultaneous co-expression of mRNAs and proteins yields mainly weak correlations. Therefore we introduce the concept of time-delayed co-expression patterns. Based on a time-course dataset, we obtain a high fraction of time-delayed correlations. In group (i), 67% of different transcripts are significantly correlated. At the protein level (ii), 68% of different proteins are significantly correlated. Comparison of the different molecular levels results in a 74% fraction of correlated transcript and protein levels of different genes (iii) and 56% for the same genes (iv). Furthermore, a higher fraction of protein levels (simultaneously 20% and short time-delayed 29%) is correlated than at the transcript level (10% and 18% respectively). Analysis of the dynamics of the correlation shows that correlation at the transcript level is largely passed to the protein level. In contrast, specific co-expression patterns are changed in multiple ways.

Conclusions

Our analysis reveals that the regulation of transcription and translation contains a time-delayed component. The correlation at the protein level is more synchronous or delayed by shorter time than those at the transcript level. This supports the hypothesis that a higher degree of direct physical interactions require a higher synchronicity between the interacting partners. The conservation of correlation between the transcript level (i) and the protein level (ii) sheds light on the processes underlying transcription, translation and regulation. A future investigation of the conditions of conservation will give comprehensive insights in the complexity of the regulatory mechanisms.

A scale space approach for unsupervised feature selection in mass spectra classification for ovarian cancer detection

BACKGROUND

Mass spectrometry spectra, widely used in proteomics studies as a screening tool for protein profiling and to detect discriminatory signals, are high dimensional data. A large number of local maxima (a.k.a. peaks) have to be analyzed as part of computational pipelines aimed at the realization of efficient predictive and screening protocols. With this kind of data dimensions and samples size the risk of over-fitting and selection bias is pervasive. Therefore the development of bio-informatics methods based on unsupervised feature extraction can lead to general tools which can be applied to several fields of predictive proteomics.

RESULTS

We propose a method for feature selection and extraction grounded on the theory of multi-scale spaces for high resolution spectra derived from analysis of serum. Then we use support vector machines for classification. In particular we use a database containing 216 samples spectra divided in 115 cancer and 91 control samples. The overall accuracy averaged over a large cross validation study is 98.18. The area under the ROC curve of the best selected model is 0.9962.

CONCLUSION

We improved previous known results on the problem on the same data, with the advantage that the proposed method has an unsupervised feature selection phase. All the developed code, as MATLAB scripts, can be downloaded from http://medeaserver.isa.cnr.it/dacierno/spectracode.htm.

Predicting neuroendocrine tumor (carcinoid) neoplasia using gene expression profiling and supervised machine learning

BACKGROUND

A more accurate taxonomy of small intestinal (SI) neuroendocrine tumors (NETs) is necessary to accurately predict tumor behavior and prognosis and to define therapeutic strategy. In this study, the authors identified a panel of such markers that have been implicated in tumorigenicity, metastasis, and hormone production and hypothesized that transcript levels of the genes melanoma antigen family D2 (MAGE-D2), metastasis-associated 1 (MTA1), nucleosome assembly protein 1-like (NAP1L1), Ki-67 (a marker of proliferation), survivin, frizzled homolog 7 (FZD7), the Kiss1 metastasis suppressor (Kiss1), neuropilin 2 (NRP2), and chromogranin A (CgA) could be used to define primary SI NETs and to predict the development of metastases.

METHODS

Seventy-three clinically and World Health Organization pathologically classified NET samples (primary tumor, n = 44 samples; liver metastases, n = 29 samples) and 30 normal human enterochromaffin (EC) cell preparations were analyzed using real-time polymerase chain reaction. Transcript levels were normalized to 3 NET housekeeping genes (asparagine-linked glycosylation 9 or ALG9, transcription factor CP2 or TFCP2, and zinc finger protein 410 or ZNF410) using geNorm analysis. A predictive gene-based model was constructed using supervised learning algorithms from the transcript expression levels.

RESULTS

Primary SI NETs could be differentiated from normal human EC cell preparations with 100% specificity and 92% sensitivity. Well differentiated NETs (WDNETs), well differentiated neuroendocrine carcinomas, and poorly differentiated NETs (PDNETs) were classified with a specificity of 78%, 78%, and 71%, respectively; whereas poorly differentiated neuroendocrine carcinomas were misclassified as either WDNETs or PDNETs. Metastases were predicted in all cases with 100% sensitivity and specificity.

CONCLUSIONS

The current results indicated that gene expression profiling and supervised machine learning can be used to classify SI NET subtypes and accurately predict metastasis. The authors believe that the application of this technique will facilitate accurate molecular pathologic delineation of NET disease, better define its extent, facilitate the assessment of prognosis, and provide a guide for the identification of appropriate strategies for individualized patient treatment.

Microarray analyses of peripheral blood cells identifies unique gene expression signature in psoriatic arthritis

Psoriatic arthritis (PsA) is a chronic and erosive form of arthritis of unknown cause. We aimed to characterize the PsA phenotype using gene expression profiling and comparing it with healthy control subjects and patients rheumatoid arthritis (RA). Peripheral blood cells (PBCs) of 19 patients with active PsA and 19 age- and sex-matched control subjects were used in the analyses of PsA, with blood samples collected in PaxGene tubes. A significant alteration in the pattern of expression of 313 genes was noted in the PBCs of PsA patients on Affymetrix U133A arrays: 257 genes were expressed at reduced levels in PsA, and 56 genes were expressed at increased levels, compared with controls. Downregulated genes tended to cluster to certain chromosomal regions, including those containing the psoriasis susceptibility loci PSORS1 and PSORS2. Among the genes with the most significantly reduced expression were those involved in downregulation or suppression of innate and acquired immune responses, such as SIGIRR, STAT3, SHP1, IKBKB, IL-11RA, and TCF7, suggesting inappropriate control that favors proin-flammatory responses. Several members of the MAPK signaling pathway and tumor suppressor genes showed reduced expression. Three proinflammatory genes--S100A8, S100A12, and thioredoxin--showed increased expression. Logistic regression and recursive partitioning analysis determined that one gene, nucleoporin 62 kDa, could correctly classify all controls and 94.7% of the PsA patients. Using a dataset of 48 RA samples for comparison, the combination of two genes, MAP3K3 followed by CACNA1S, was enough to correctly classify all RA and PsA patients. Thus, PBC gene expression profiling identified a gene expression signature that differentiated PsA from RA, and PsA from controls. Several novel genes were differentially expressed in PsA and may prove to be diagnostic biomarkers or serve as new targets for the development of therapies.

An integrated method for cancer classification and rule extraction from microarray data

Different microarray techniques recently have been successfully used to investigate useful information for cancer diagnosis at the gene expression level due to their ability to measure thousands of gene expression levels in a massively parallel way. One important issue is to improve classification performance of microarray data. However, it would be ideal that influential genes and even interpretable rules can be explored at the same time to offer biological insight.

Introducing the concepts of system design in software engineering, this paper has presented an integrated and effective method (named X-AI) for accurate cancer classification and the acquisition of knowledge from DNA microarray data. This method included a feature selector to systematically extract the relative important genes so as to reduce the dimension and retain as much as possible of the class discriminatory information. Next, diagonal quadratic discriminant analysis (DQDA) was combined to classify tumors, and generalized rule induction (GRI) was integrated to establish association rules which can give an understanding of the relationships between cancer classes and related genes.

Two non-redundant datasets of acute leukemia were used to validate the proposed X-AI, showing significantly high accuracy for discriminating different classes. On the other hand, I have presented the abilities of X-AI to extract relevant genes, as well as to develop interpretable rules. Further, a web server has been established for cancer classification and it is freely available at .

Serum protein signatures detect early radiographic osteoarthritis

OBJECTIVE

To test the hypothesis that early knee and hand osteoarthritis (OA) development is characterized by detectable changes in serum proteins relevant to inflammation, cell growth, activation, and metabolism several years before OA becomes radiographically evident.

METHODS

Using microarray platforms that simultaneously test 169 proteins relevant to inflammation, cell growth, activation and metabolism, we conducted a case-control study nested within the Baltimore Longitudinal Study of Aging (BLSA). Subjects included 22 incident cases of OA and 66 age-, sex- and body mass index (BMI)-matched controls. Serum samples tested were obtained at the time of radiographic classification as either case or control, and up to 10 years earlier at a time when all participants were free of radiographic OA. Proteins with mean signal intensities fourfold higher than background were compared between cases and controls using multivariate techniques.

RESULTS

Sixteen proteins were different between OA cases compared to controls. Four of these proteins [matrix metalloproteinase (MMP)-7, interleukin (IL)-15, plasminogen activator inhibitor (PAI)-1 and soluble vascular adhesion protein (sVAP)-1] were already different in samples obtained 10 years before radiographic classification and remained different at the time of diagnosis. Six additional proteins were only associated with subsequent OA development and not with established OA.

CONCLUSIONS

Changes in serum proteins implicated in matrix degradation, cell activation, inflammation and bone collagen degradation products accompany early OA development and can precede radiographic detection by several years.

Identification of cancer-associated gene clusters and genes via clustering penalization

Identification of genes associated with cancer development and progression using microarray data is challenging because of the high dimensionality and cluster structure of gene expressions. Here the clusters are composed of multiple genes with coordinated biological functions and/or correlated expressions. In this article, we first propose a hybrid approach for clustering gene expressions. The hybrid approach uses both pathological pathway information and correlations of gene expressions. We propose using the group bridge, a novel clustering penalization approach, for analysis of cancer microarray data. The group bridge approach explicitly accounts for the cluster structure of gene expressions, and is capable of selecting gene clusters and genes within those selected clusters that are associated with cancer. We also develop an iterative algorithm for computing the group bridge estimator. Analysis of three cancer microarray datasets shows that the proposed approach can identify biologically meaningful gene clusters and genes within those identified clusters.

Squalene epoxidase, located on chromosome 8q24.1, is upregulated in 8q+ breast cancer and indicates poor clinical outcome in stage I and II disease

Gains of chromosomes 7p and 8q are associated with poor prognosis among oestrogen receptor-positive (ER+) stage I/II breast cancer. To identify transcriptional changes associated with this breast cancer subtype, we applied suppression subtractive hybridisation method to analyse differentially expressed genes among six breast tumours with and without chromosomal 7p and 8q gains. Identified mRNAs were validated by real-time RT–PCR in tissue samples obtained from 186 patients with stage I/II breast cancer. Advanced statistical methods were applied to identify associations of mRNA expression with distant metastasis-free survival (DMFS). mRNA expression of the key enzyme of cholesterol biosynthesis, squalene epoxidase (SQLE, chromosomal location 8q24.1), was associated with ER+ 7p+/8q+ breast cancer. Distant metastasis-free survival in stage I/II breast cancer cases was significantly inversely related to SQLE mRNA in multivariate Cox analysis (P<0.001) in two independent patient cohorts of 160 patients each. The clinically favourable group associated with a low SQLE mRNA expression could be further divided by mRNA expression levels of the oestrogen-regulated zinc transporter LIV-1. The data strongly support that SQLE mRNA expression might indicate high-risk ER+ stage I/II breast cancers. Further studies on tumour tissue from standardised treated patients, for example with tamoxifen, may validate the role of SQLE as a novel diagnostic parameter for ER+ early stage breast cancers.

Molecular diagnosis of human cancer type by gene expression profiles and independent component analysis

The precise diagnosis of cancer type based on microarray data is of particular importance and is also a challenging task. We have devised a novel pattern recognition procedure based on independent component analysis (ICA). Different from the conventional cancer classification methods, which are limited in their clinical applicability of cancer diagnosis, our method extracts explicitly, by ICA algorithm, a set of specific diagnostic patterns of normal and tumor tissues corresponding to a set of biomarkers for clinical use. We validated our procedure with the colon and prostate cancer data sets and achieved good diagnosis (>90%) on the data sets studied here. This technique is also suitable for the identification of diagnostic expression patterns for other human cancers and demonstrates the feasibility of simple and accurate molecular cancer diagnostics for clinical implementation.

Discovering high-order patterns of gene expression levels

This paper reports the discovery of statistically significant association patterns of gene expression levels from microarray data. By association patterns, we mean certain gene expression intensity intervals having statistically significant associations among themselves and with the tissue classes, such as cancerous and normal tissues. We describe how the significance of the associations among gene expression levels can be evaluated using a statistical measure in an objective manner. If an association is found to be significant based on the measure, we say that it is statistically significant. Given a gene expression data set, we first cluster the entire gene pool comprising all the genes into groups by optimizing the correlation (or more precisely, interdependence) among the gene expression levels within gene groups. From each group, we select one or several genes that are most correlated with other genes within that group to form a smaller gene pool. This gene pool then constitutes the most representative genes from the original pool. Our pattern discovery algorithm is then used, for the first time, to discover the significant association patterns of gene expression levels among the genes from the small pool. With our method, it is more effective to discover and express the associations in terms of their intensity intervals. Hence, we discretize each gene expression levels into intervals maximizing the interdependence between the gene expression and the tissue classes. From this data set of gene expression intervals, we discover the association patterns representing statistically significant associations, some positively and some negatively, with different tissue classes. We apply our pattern discovery methodology to the colon-cancer microarray gene expression data set. It consists of 2000 genes and 62 samples taken from colon cancer or normal subjects. The statistically significant combinations of gene expression levels that repress or activate colon cancer are revealed in the colon-cancer data set. The discovered association patterns are ranked according to their statistical significance and displayed for interpretation and further analysis.

Prediction of relapse in paediatric pre-B acute lymphoblastic leukaemia using a three-gene risk index

Despite high cure rates 25% of children with acute lymphoblastic leukaemia (ALL) relapse and have dismal outcome. Crucially, many are currently stratified as standard risk (SR) and additional markers to improve patient stratification are required. Here we have used diagnostic bone marrow specimens from 101 children with pre-B ALL to examine the use of gene expression profiles (GEP) as predictors of long-term clinical outcome. Patients were divided into two cohorts for model development and validation based on availability of specimen material. Initially, GEP from 55 patients with sufficient material were analysed using HG-U133A microarrays, identifying an 18-gene classifier (GC) that was more predictive of outcome than conventional prognostic parameters. After feature selection and validation of expression levels by quantitative reverse transcription polymerase chain reaction (qRT-PCR), a three-gene qRT-PCR risk index [glutamine synthetase (GLUL), ornithine decarboxylase antizyme inhibitor (AZIN), immunoglobulin J chain (IGJ)] was developed that predicted outcome with an accuracy of 89% in the array cohort and 87% in the independent validation cohort. The data demonstrate the feasibility of using GEP to improve risk stratification in childhood ALL. This is particularly important for the identification of patients destined to relapse despite their current stratification as SR, as more intensive front-line treatment options for these individuals are already available.

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.

A new method for class prediction based on signed-rank algorithms applied to Affymetrix microarray experiments

Background

The huge amount of data generated by DNA chips is a powerful basis to classify various pathologies. However, constant evolution of microarray technology makes it difficult to mix data from different chip types for class prediction of limited sample populations. Affymetrix^® technology provides both a quantitative fluorescence signal and a decision (detection call: absent or present) based on signed-rank algorithms applied to several hybridization repeats of each gene, with a per-chip normalization. We developed a new prediction method for class belonging based on the detection call only from recent Affymetrix chip type. Biological data were obtained by hybridization on U133A, U133B and U133Plus 2.0 microarrays of purified normal B cells and cells from three independent groups of multiple myeloma (MM) patients.

Results

After a call-based data reduction step to filter out non class-discriminative probe sets, the gene list obtained was reduced to a predictor with correction for multiple testing by iterative deletion of probe sets that sequentially improve inter-class comparisons and their significance. The error rate of the method was determined using leave-one-out and 5-fold cross-validation. It was successfully applied to (i) determine a sex predictor with the normal donor group classifying gender with no error in all patient groups except for male MM samples with a Y chromosome deletion, (ii) predict the immunoglobulin light and heavy chains expressed by the malignant myeloma clones of the validation group and (iii) predict sex, light and heavy chain nature for every new patient. Finally, this method was shown powerful when compared to the popular classification method Prediction Analysis of Microarray (PAM).

Conclusion

This normalization-free method is routinely used for quality control and correction of collection errors in patient reports to clinicians. It can be easily extended to multiple class prediction suitable with clinical groups, and looks particularly promising through international cooperative projects like the "Microarray Quality Control project of US FDA" MAQC as a predictive classifier for diagnostic, prognostic and response to treatment. Finally, it can be used as a powerful tool to mine published data generated on Affymetrix systems and more generally classify samples with binary feature values.