Feature Selection and Classification of MAQC-II Breast Cancer and Multiple Myeloma Microarray Gene Expression Data

Microarray data has a high dimension of variables but available datasets usually have only a small number of samples, thereby making the study of such datasets interesting and challenging. In the task of analyzing microarray data for the purpose of, e.g., predicting gene-disease association, feature selection is very important because it provides a way to handle the high dimensionality by exploiting information redundancy induced by associations among genetic markers. Judicious feature selection in microarray data analysis can result in significant reduction of cost while maintaining or improving the classification or prediction accuracy of learning machines that are employed to sort out the datasets. In this paper, we propose a gene selection method called Recursive Feature Addition (RFA), which combines supervised learning and statistical similarity measures. We compare our method with the following gene selection methods:

• Support Vector Machine Recursive Feature Elimination (SVMRFE)
• Leave-One-Out Calculation Sequential Forward Selection (LOOCSFS)
• Gradient based Leave-one-out Gene Selection (GLGS)

To evaluate the performance of these gene selection methods, we employ several popular learning classifiers on the MicroArray Quality Control phase II on predictive modeling (MAQC-II) breast cancer dataset and the MAQC-II multiple myeloma dataset. Experimental results show that gene selection is strictly paired with learning classifier. Overall, our approach outperforms other compared methods. The biological functional analysis based on the MAQC-II breast cancer dataset convinced us to apply our method for phenotype prediction. Additionally, learning classifiers also play important roles in the classification of microarray data and our experimental results indicate that the Nearest Mean Scale Classifier (NMSC) is a good choice due to its prediction reliability and its stability across the three performance measurements: Testing accuracy, MCC values, and AUC errors.     

Minimising immunohistochemical false negative ER classification using a complementary 23 gene expression signature of ER status

Background

Expression of the oestrogen receptor (ER) in breast cancer predicts benefit from endocrine therapy. Minimising the frequency of false negative ER status classification is essential to identify all patients with ER positive breast cancers who should be offered endocrine therapies in order to improve clinical outcome. In routine oncological practice ER status is determined by semi-quantitative methods such as immunohistochemistry (IHC) or other immunoassays in which the ER expression level is compared to an empirical threshold[1], [2]. The clinical relevance of gene expression-based ER subtypes as compared to IHC-based determination has not been systematically evaluated. Here we attempt to reduce the frequency of false negative ER status classification using two gene expression approaches and compare these methods to IHC based ER status in terms of predictive and prognostic concordance with clinical outcome.

Methodology/Principal Findings

Firstly, ER status was discriminated by fitting the bimodal expression of ESR1 to a mixed Gaussian model. The discriminative power of ESR1 suggested bimodal expression as an efficient way to stratify breast cancer; therefore we identified a set of genes whose expression was both strongly bimodal, mimicking ESR expression status, and highly expressed in breast epithelial cell lines, to derive a 23-gene ER expression signature-based classifier. We assessed our classifiers in seven published breast cancer cohorts by comparing the gene expression-based ER status to IHC-based ER status as a predictor of clinical outcome in both untreated and tamoxifen treated cohorts. In untreated breast cancer cohorts, the 23 gene signature-based ER status provided significantly improved prognostic power compared to IHC-based ER status (P = 0.006). In tamoxifen-treated cohorts, the 23 gene ER expression signature predicted clinical outcome (HR = 2.20, P = 0.00035). These complementary ER signature-based strategies estimated that between 15.1% and 21.8% patients of IHC-based negative ER status would be classified with ER positive breast cancer.

Conclusion/Significance

Expression-based ER status classification may complement IHC to minimise false negative ER status classification and optimise patient stratification for endocrine therapies.     

Eigengene-based linear discriminant model for tumor classification using gene expression microarray data

MOTIVATION: The nearest shrunken centroids classifier has become a popular algorithm in tumor classification problems using gene expression microarray data. Feature selection is an embedded part of the method to select top-ranking genes based on a univariate distance statistic calculated for each gene individually. The univariate statistics summarize gene expression profiles outside of the gene co-regulation network context, leading to redundant information being included in the selection procedure. RESULTS: We propose an Eigengene-based Linear Discriminant Analysis (ELDA) to address gene selection in a multivariate framework. The algorithm uses a modified rotated Spectral Decomposition (SpD) technique to select 'hub' genes that associate with the most important eigenvectors. Using three benchmark cancer microarray datasets, we show that ELDA selects the most characteristic genes, leading to substantially smaller classifiers than the univariate feature selection based analogues. The resulting de-correlated expression profiles make the gene-wise independence assumption more realistic and applicable for the shrunken centroids classifier and other diagonal linear discriminant type of models. Our algorithm further incorporates a misclassification cost matrix, allowing differential penalization of one type of error over another. In the breast cancer data, we show false negative prognosis can be controlled via a cost-adjusted discriminant function. AVAILABILITY: R code for the ELDA algorithm is available from author upon request.     

Mixture models for detecting differentially expressed genes in microarrays

An important and common problem in microarray experiments is the detection of genes that are differentially expressed in a given number of classes. As this problem concerns the selection of significant genes from a large pool of candidate genes, it needs to be carried out within the framework of multiple hypothesis testing. In this paper, we focus on the use of mixture models to handle the multiplicity issue. With this approach, a measure of the local FDR (false discovery rate) is provided for each gene. An attractive feature of the mixture model approach is that it provides a framework for the estimation of the prior probability that a gene is not differentially expressed, and this probability can subsequently be used in forming a decision rule. The rule can also be formed to take the false negative rate into account. We apply this approach to a well-known publicly available data set on breast cancer, and discuss our findings with reference to other approaches.     

Distinct genes related to drug response identified in ER positive and ER negative breast cancer cell lines

Breast cancer patients have different responses to chemotherapeutic treatments. Genes associated with drug response can provide insight to understand the mechanisms of drug resistance, identify promising therapeutic opportunities, and facilitate personalized treatment. Estrogen receptor (ER) positive and ER negative breast cancer have distinct clinical behavior and molecular properties. However, to date, few studies have rigorously assessed drug response genes in them. In this study, our goal was to systematically identify genes associated with multidrug response in ER positive and ER negative breast cancer cell lines. We tested 27 human breast cell lines for response to seven chemotherapeutic agents (cyclophosphamide, docetaxel, doxorubicin, epirubicin, fluorouracil, gemcitabine, and paclitaxel). We integrated publicly available gene expression profiles of these cell lines with their in vitro drug response patterns, then applied meta-analysis to identify genes related to multidrug response in ER positive and ER negative cells separately. One hundred eighty-eight genes were identified as related to multidrug response in ER positive and 32 genes in ER negative breast cell lines. Of these, only three genes (DBI, TOP2A, and PMVK) were common to both cell types. TOP2A was positively associated with drug response, and DBI was negatively associated with drug response. Interestingly, PMVK was positively associated with drug response in ER positive cells and negatively in ER negative cells. Functional analysis showed that while cell cycle affects drug response in both ER positive and negative cells, most biological processes that are involved in drug response are distinct. A number of signaling pathways that are uniquely enriched in ER positive cells have complex cross talk with ER signaling, while in ER negative cells, enriched pathways are related to metabolic functions. Taken together, our analysis indicates that distinct mechanisms are involved in multidrug response in ER positive and ER negative breast cells.     

A method for predicting disease subtypes in presence of misclassification among training samples using gene expression: application to human breast cancer

MOTIVATION: An accurate diagnostic and prediction will not be achieved unless the disease subtype status for every training sample used in the supervised learning step is accurately known. Such an assumption requires the existence of a perfect tool for disease diagnostic and classification, which is seldom available in the majority of the cases. Thus, the supervised learning step has to be conducted with a statistical model that contemplates and handles potential mislabeling in the input data. RESULTS: A procedure for handling potential mislabeling among training samples in the prediction of disease subtypes using gene expression data was proposed. A real data-based simulation study about the estrogen receptor status (ER+/ER-) of breast cancer patients was conducted. The results demonstrated that when 1-4 training samples (N = 30) were artificially mislabeled, the proposed method was able not only in correcting the ER status of mislabeled training samples but also more importantly in predicting the ER status of validation samples as well as using 'true' training data.     

吲哚菁绿联合亚甲蓝在乳腺癌前哨淋巴结活检中的临床应用

目的:探讨吲哚菁绿(indocyanine green,ICG)联合亚甲蓝在乳腺癌前哨淋巴结活检(sentinel lymph node biopsy,SLNB)中的临床应用价值。方法:共入组39例乳腺癌患者,在光学分子影像手术导航系统(surgical navigation system,SNS)的引导下,使用ICG联合亚甲蓝实施SLNB。快速冰冻病理证实前哨淋巴结(sentinel lymph node,SLN)转移者,即刻行腋窝淋巴结清扫(axillary lymph node dissection,ALND);SLN阴性者免于ALND。结果:联合法实施SLNB检出率95%,成功检出SLN125个,平均检出3.2个SLN,准确率95.0%,灵敏度100%,假阴性率0%;蓝染法检出率87.2%,成功检出SLN71个,平均检出1.8个SLN,准确率89.7%,灵敏度93.7%,假阴性率为6.3%;统计结果采用x2检验,检验结果具有统计学意义。结论:ICG联合亚甲蓝实施乳腺癌SLNB成功率高,假阴性率低,临床效果不亚于亚甲蓝,是实施SLNB的新方法。     

Feature selection based on empirical-risk function to detect lesions in vascular computed tomography

ObjectiveThe overall goal of the study is to detect coronary artery lesions regardless their nature, calcified or hypo-dense. To avoid explicit modelling of heterogeneous lesions, we adopted an approach based on machine learning and using unsupervised or semi-supervised classifiers. The success of the classifiers based on machine learning strongly depends on the appropriate choice of features differentiating between lesions and regular appearance. The specific goal of this article is to propose a novel strategy devised to select the best feature set for the classifiers used, out of a given set of candidate features.Materials and methodsThe features are calculated in image planes orthogonal to the artery centerline, and the classifier assigns to each of these cross-sections a label “healthy” or “diseased”. The contribution of this article is a feature-selection strategy based on the empirical risk function that is used as a criterion in the initial feature ranking and in the selection process itself. We have assessed this strategy in association with two classifiers based on the density-level detection approach that seeks outliers from the distribution corresponding to the regular appearance. The method was evaluated using a total of 13,687 cross-sections extracted from 53 coronary arteries in 15 patients.ResultsUsing the feature subset selected by the risk-based strategy, balanced error rates achieved by the unsupervised and semi-supervised classifiers respectively were equal to 13.5% and 15.4%. These results were substantially better than the rates achieved using feature subsets selected by supervised strategies. The unsupervised and semi-supervised methods also outperformed supervised classifiers using feature subsets selected by the corresponding supervised strategies.DiscussionSupervised methods require large data sets annotated by experts, both to select the features and to train the classifiers, and collecting these annotations is time-consuming. With these methods, lesions whose appearance differs from the training data may remain undetected. Lesion-detection problem is highly imbalanced, since healthy cross-sections usually are much more numerous than the diseased ones. Training the classifiers based on the density-level detection approach needs a small number of annotations or no annotations at all. The same annotations are sufficient to compute the empirical risk and to perform the selection. Therefore, our strategy associated with an unsupervised or semi-supervised classifier requires a considerably smaller number of annotations as compared to conventional supervised selection strategies. The approach proposed is also better suited for highly imbalanced problems and can detect lesions differing from the training set.ConclusionThe risk-based selection strategy, associated with classifiers using the density-level detection approach, outperformed other strategies and classifiers when used to detect coronary artery lesions. It is well suited for highly imbalanced problems, where the lesions are represented as low-density regions of the feature space, and it can be used in other anomaly detection problems interpretable as a binary classification problem where the empirical risk can be calculated.     

Expression of oestrogen receptor-beta in invasive breast tumours

Steroid hormone receptors are used routinely to predict endocrine responsiveness in patients with breast cancer. Two oestrogen receptors (ERs): ER alpha and ER beta have been identified. Although ER alpha and ER beta genes share a large degree of homology, it is generally thought that their distribution and function are substantially different in many tissues. Both of them may be expressed in normal and neoplastic tissues of the breast. While much is known about ER alpha, the role of ER beta is still undefined, especially at the protein level. Recent development of reliable antibodies to ER beta has provided opportunity to test immunohistochemical reactions detecting ER beta in archival breast tumours. The aim of our study was to learn more about the cellular mechanisms underlying the relationship of ER beta and progesterone receptor (PR) in breast cancer tissues, discriminating between hormone-dependent and hormone-independent tumours. ER alpha and PR content of tumour tissues of 154 patients with breast cancer were tested by in situ indirect immunohistochemical method parallel with ligand binding biochemical assay. ER beta was detected in 8 ER alpha-/PR+ breast carcinomas by immunohistochemical method too. Steroid hormone receptor content was analysed comparing to the histologic type and grading of the tumours. CONCLUSIONS: A considerable part of breast carcinomas belongs to the ER+/PR+ and ER-/PR- groups. About 1-2% of the tumours is expected to be ER alpha-/ER beta+/PR+ type. In such cases ER alpha negative reaction together with PR positivity can signal the necessity of the immunohistochemical detection of ER beta in routine histopathological practice, presenting the precise steroid hormone receptor status for the most effective endocrine therapy of the patients.     

Estrogen receptor-alpha 36 mediates the anti-apoptotic effect of estradiol in triple negative breast cancer cells via a membrane-associated mechanism

17β-Estradiol can promote the growth and development of several estrogen receptor (ER)-negative breast cancers. The effects are rapid and non-genomic, suggesting that a membrane-associated ER is involved. ERα36 has been shown to mediate rapid, non-genomic, membrane-associated effects of 17β-estradiol in several cancer cell lines, including triple negative HCC38 breast cancer cells. Moreover, the effect is anti-apoptotic. The aim of this study was to determine if ERα36 mediates this anti-apoptotic effect, and to elucidate the mechanism involved. Taxol was used to induce apoptosis in HCC38 cells, and the effect of 17β-estradiol pre-treatment was determined. Antibodies to ERα36, signal pathway inhibitors, ERα36 deletion mutants, and ERα36-silencing were used prior to these treatments to determine the role of ERα36 in these effects and to determine which signaling molecules were involved. We found that the anti-apoptotic effect of 17β-estradiol in HCC38 breast cancer cells is in fact mediated by membrane-associated ERα36. We also showed that this signaling occurs through a pathway that requires PLD, LPA, and PI3K; Gαs and calcium signaling may also be involved. In addition, dynamic palmitoylation is required for the membrane-associated effect of 17β-estradiol. Exon 9 of ERα36, a unique exon to ERα36 not found in other identified splice variants of ERα with previously unknown function, is necessary for these effects. This study provides a working model for a mechanism by which estradiol promotes anti-apoptosis through membrane-associated ERα36, suggesting that ERα36 may be a potential membrane target for drug design against breast cancer, particularly triple negative breast cancer.     

Development, characterization, and applications of a novel estrogen receptor beta monoclonal antibody

The role of estrogen receptor alpha (ERα) in breast cancer has been studied extensively, and its protein expression is prognostic and a primary determinant of endocrine sensitivity. However, much less is known about the role of ERβ and its relevance remains unclear due to the publication of conflicting reports. Here, we provide evidence that much of this controversy may be explained by variability in antibody sensitivity and specificity and describe the development, characterization, and potential applications of a novel monoclonal antibody targeting full-length human ERβ and its splice variant forms. Specifically, we demonstrate that a number of commercially available ERβ antibodies are insensitive for ERβ and exhibit significant cross-reaction with ERα. However, our newly developed MC10 ERβ antibody is shown to be highly specific and sensitive for detection of full-length ERβ and its variant forms. Strong and variable staining patterns for endogenous levels of ERβ protein were detected in normal human tissues and breast tumors using the MC10 antibody. Importantly, ERβ was shown to be expressed in a limited cohort of both ERα positive and ERα negative breast tumors. Taken together, these data demonstrate that the use of poorly validated ERβ antibodies is likely to explain much of the controversy in the field with regard to the biological relevance of ERβ in breast cancer. The use of the MC10 antibody, in combination with highly specific antibodies targeting only full-length ERβ, is likely to provide additional discriminatory features in breast cancers that may be useful in predicting response to therapy.     

Estrogen receptor-β activation in combination with letrozole blocks the growth of breast cancer tumors resistant to letrozole therapy

Treatment with anti-estrogens or aromatase inhibitors (AI) is the main therapeutic strategy used against estrogen receptor ERα-positive breast cancer. Resistance to these therapies presents a major challenge in the management of breast cancer. Little is known about ERβ in breast carcinogenesis. Our aim in this study is to examine potential novel strategies utilizing ERβ activity to overcome AI resistance. We provide evidence that ERβ agonist can reduce the growth of AI-resistant breast cancer cells. Our data further confirm that therapeutic activation of ERβ by DPN, an ERβ agonist, blocks letrozole-resistant tumor growth in a xenograft model. Interestingly, DPN exerted tumor growth inhibition only in the presence of the AI letrozole, suggesting that combination therapy including ERβ activators and AI may be used in the clinical setting treating AI resistant breast cancer. An increase in ERβ levels, with diminished ERα/ERβ ratio, was observed in the tumors from mice treated with DPN/letrozole combination compared to single agents and control. Decreased Cyclin D1 and increased CyclinD1/CDK inhibitors p21 and p27 levels in DPN/letrozole treated tumors were observed, suggesting that the combination treatment may inhibit tumor growth by blocking G1/S phase cell cycle progression. Our data show a decrease in MAPK phosphorylation levels without affecting total levels. In addition to providing evidence suggesting the potential use of ERβ agonists in combination with letrozole in treating AI resistant breast cancer and prolonging sensitivity to AI, we also provide mechanistic evidence supporting the role of ERβ in altering the expression profile associated with resistance.     

ERα-negative and triple negative breast cancer: Molecular features and potential therapeutic approaches

Triple negative breast cancer (TNBC) is a type of aggressive breast cancer lacking the expression of estrogen receptors (ER), progesterone receptors (PR) and human epidermal growth factor receptor-2 (HER-2). TNBC patients account for approximately 15% of total breast cancer patients and are more prevalent among young African, African-American and Latino women patients. The currently available ER-targeted and Her-2-based therapies are not effective for treating TNBC. Recent studies have revealed a number of novel features of TNBC. In the present work, we comprehensively addressed these features and discussed potential therapeutic approaches based on these features for TNBC, with particular focus on: 1) the pathological features of TNBC/basal-like breast cancer; 2) E₂/ERβ-mediated signaling pathways; 3) G-protein coupling receptor-30/epithelial growth factor receptor (GPCR-30/EGFR) signaling pathway; 4) interactions of ERβ with breast cancer 1/2 (BRCA1/2); 5) chemokine CXCL8 and related chemokines; 6) altered microRNA signatures and suppression of ERα expression/ERα-signaling by micro-RNAs; 7) altered expression of several pro-oncongenic and tumor suppressor proteins; and 8) genotoxic effects caused by oxidative estrogen metabolites. Gaining better insights into these molecular pathways in TNBC may lead to identification of novel biomarkers and targets for development of diagnostic and therapeutic approaches for prevention and treatment of TNBC.     

Adenovirus-directed expression of dominant negative estrogen receptor induces apoptosis in breast cancer cells and regression of tumors in nude mice.

BACKGROUND: Estrogen receptors (ER) are expressed in about two thirds of human breast cancer, and are an important pharmacological target for treatment of these tumors. Dominant negative forms of the ER have been suggested as an alternative method to disrupt ER function. In this study, we examined the effect of dominant negative ER mutants (ER1-536 and L540Q) on ER-positive breast cancer cells in vitro and in vivo. MATERIALS AND METHODS: ER-positive T47D breast cancer cells were infected with adenoviral vectors expressing ER1-536 and L540Q to examine the effects of the mutants on gene expression and cell growth. Adenoviral vectors containing the wild type ER (AdwtER) and beta-galactosidase gene (AdGal) were used as controls. RESULTS: Ad1-536 or AdL540Q infection inhibited T47D cell growth and induced apoptosis, increasing Bax protein and phosphorylation of p38 mitogen-activated-protein kinase (MAPK). Consistent with the apoptotic effects in vitro, pre-infection of T47D cells with Ad1-536 or AdL540Q inhibited tumor formation when these cells were introduced into nude mice. In addition, injection of Ad1-536 and AdL540Q into pre-established T47D tumors induced tumor regression. Apoptosis, in conjunction with the activation of caspase-3 and phosphorylation of p38 MAPK, was detected in the shrinking tumors. Overexpression of wild-type ER by AdwtER infection also produced antiproliferative and apoptotic effects, but to a lesser extent than the ER1-536 and L540Q mutants. CONCLUSIONS: These results indicate that dominant negative ER mutants have the potential to induce apoptosis of T47D cells and regression of tumors. The delivery of dominant negative ERs by adenoviral vectors may provide a useful tool for targeted therapy of ER-positive breast cancer.     

Coupling proteomics and transcriptomics in the quest of subtype‐specific proteins in breast cancer

Breast‐cancer subtypes present with distinct clinical characteristics. Therefore, characterization of subtype‐specific proteins may augment the development of targeted therapies and prognostic biomarkers. To address this issue, MS‐based secretome analysis of eight breast cancer cell lines, corresponding to the three main breast cancer subtypes was performed. More than 5200 non‐redundant proteins were identified with 23, four, and four proteins identified uniquely in basal, HER2‐neu‐amplified, and luminal breast cancer cells, respectively. An in silico mRNA analysis using publicly available breast cancer tissue microarray data was carried out as a preliminary verification step. In particular, the expression profiles of 15 out of 28 proteins included in the microarray (from a total of 31 in our subtype‐specific signature) showed significant correlation with estrogen receptor (ER) expression. A MS‐based analysis of breast cancer tissues was undertaken to verify the results at the proteome level. Eighteen out of 31 proteins were quantified in the proteomes of ER‐positive and ER‐negative breast cancer tissues. Survival analysis using microarray data was performed to examine the prognostic potential of these selected candidates. Three proteins correlated with ER status at both mRNA and protein levels: ABAT, PDZK1, and PTX3, with the former showing significant prognostic potential.     

Predicting Classifier Performance with Limited Training Data: Applications to Computer-Aided Diagnosis in Breast and Prostate Cancer

Clinical trials increasingly employ medical imaging data in conjunction with supervised classifiers, where the latter require large amounts of training data to accurately model the system. Yet, a classifier selected at the start of the trial based on smaller and more accessible datasets may yield inaccurate and unstable classification performance. In this paper, we aim to address two common concerns in classifier selection for clinical trials: (1) predicting expected classifier performance for large datasets based on error rates calculated from smaller datasets and (2) the selection of appropriate classifiers based on expected performance for larger datasets. We present a framework for comparative evaluation of classifiers using only limited amounts of training data by using random repeated sampling (RRS) in conjunction with a cross-validation sampling strategy. Extrapolated error rates are subsequently validated via comparison with leave-one-out cross-validation performed on a larger dataset. The ability to predict error rates as dataset size increases is demonstrated on both synthetic data as well as three different computational imaging tasks: detecting cancerous image regions in prostate histopathology, differentiating high and low grade cancer in breast histopathology, and detecting cancerous metavoxels in prostate magnetic resonance spectroscopy. For each task, the relationships between 3 distinct classifiers (k-nearest neighbor, naive Bayes, Support Vector Machine) are explored. Further quantitative evaluation in terms of interquartile range (IQR) suggests that our approach consistently yields error rates with lower variability (mean IQRs of 0.0070, 0.0127, and 0.0140) than a traditional RRS approach (mean IQRs of 0.0297, 0.0779, and 0.305) that does not employ cross-validation sampling for all three datasets.