Large Microarray Datasets Research Articles

NNFSRR: Nearest Neighbor Feature Selection and Redundancy Removal Method for Nearest Neighbor Search in Microarray Gene Expression Data

INTRODUCTION: Gene expression data analysis is a critical aspect of disease prediction and classification, playing a pivotal role in the field of bioinformatics and biomedical research. High-dimensional gene expression datasets hold a wealth of information, but their effective utilization is hindered by the presence of irrelevant dimensions and noise. The challenge lies in extracting meaningful features from these datasets to enhance the accuracy of disease prediction and classification while maintaining computational efficiency.
 Feature selection is a crucial step in addressing these challenges, as it aims to identify and retain only the most informative characteristics from large high-dimensional microarray datasets. In the context of microarray gene expression data, characterized by its substantial dimensionality, selecting relevant features is essential for efficient nearest neighbor search, a fundamental component of various analytical tasks in bioinformatics and data mining.
 Existing feature selection methods in high-dimensional data often face issues related to the trade-off between search accuracy and computational efficiency. This paper introduces a novel approach, the Nearest Neighbor Feature Selection with Symmetrical Uncertainty-based Redundancy Removal (NNFSRR) method, designed to enhance the classification of microarray gene expression data through feature selection. The NNFSRR method focuses on reducing the dimensionality of the dataset by identifying and removing redundant features, allowing subsequent searches to operate solely on relevant dimensions.
 OBJECTIVES: The primary goal is to evaluate the NNFSRR method's effectiveness in improving nearest neighbor search in microarray gene expression datasets by reducing dimensionality. This method utilizes Symmetrical Uncertainty-based correlation between dimensions for feature selection and aims to enhance accuracy and efficiency compared to existing methods.
 METHODS: The NNFSRR method uses Symmetrical Uncertainty to identify and remove redundant features from microarray gene expression datasets. Reduced datasets are used for nearest neighbor search, improving accuracy and efficiency. Experiments are conducted using real-world datasets, and comparisons with existing methods are made based on search time and accuracy.
 RESULTS: The NNFSRR method demonstrates improved nearest neighbor search performance, outperforming basic brute force methods and existing feature selection techniques. Selected feature sets exhibit strong class associations while minimizing feature correlations, enhancing classification precision.
 CONCLUSION: In conclusion, the NNFSRR method presents a promising approach to address the challenges posed by high-dimensional gene expression data. It effectively reduces dimensionality, improves search accuracy, and enhances the efficiency of nearest neighbor search. Our experimental results demonstrate that this method outperforms existing techniques in terms of search time and accuracy, making it a valuable tool for applications in bioinformatics, data mining, pattern recognition, and biological information retrieval. The NNFSRR method holds the potential to advance our understanding of complex biological processes and support more accurate disease prediction and classification.

Novel Feature Selection Algorithms Based on Crowding Distance and Pearson Correlation Coefficient

Feature Selection is an important phase in classification models. Feature Selection is an effective task used to decrease the dimensionality and eliminate redundant and unrelated features. In this paper, three novel algorithms for feature selection problem are proposed. The first one is a filter method, the second one is a wrapper method, and the last one is a hybrid filter method. Both the proposed algorithms use the crowding distance used in the multiobjective optimization as a new metric to assess the importance of the features. The idea behind the use of the crowding distance is that the less crowded features have great impacts on the target attribute (class), and the crowded features have generally the same impact on the class attribute. To enhance the crowded distance, a combination with other metrics will give good results. In this work, the hybrid method combines between the crowding distance and Pearson correlation coefficient to well order the importance of features. Experiments on well-known benchmark datasets including large microarray datasets have shown the effectiveness and the robustness of the proposed algorithms.

Survivin' Acute Myeloid Leukaemia-A Personalised Target for inv(16) Patients.

Despite recent advances in therapies including immunotherapy, patients with acute myeloid leukaemia (AML) still experience relatively poor survival rates. The Inhibition of Apoptosis (IAP) family member, survivin, also known by its gene and protein name, Baculoviral IAP Repeat Containing 5 (BIRC5), remains one of the most frequently expressed antigens across AML subtypes. To better understand its potential to act as a target for immunotherapy and a biomarker for AML survival, we examined the protein and pathways that BIRC5 interacts with using the Kyoto Encyclopedia of Genes and Genomes (KEGG), search tool for recurring instances of neighbouring genes (STRING), WEB-based Gene Set Analysis Toolkit, Bloodspot and performed a comprehensive literature review. We then analysed data from gene expression studies. These included 312 AML samples in the Microarray Innovations In Leukemia (MILE) dataset. We found a trend between above median levels of BIRC5 being associated with improved overall survival (OS) but this did not reach statistical significance (p = 0.077, Log-Rank). There was some evidence of a beneficial effect in adjusted analyses where above median levels of BIRC5 were shown to be associated with improved OS (p = 0.001) including in Core Binding Factor (CBF) patients (p = 0.03). Above median levels of BIRC5 transcript were associated with improved relapse free survival (p < 0.0001). Utilisation of a second large cDNA microarray dataset including 306 AML cases, again showed no correlation between BIRC5 levels and OS, but high expression levels of BIRC5 correlated with worse survival in inv(16) patients (p = 0.077) which was highly significant when datasets A and B were combined (p = 0.001). In addition, decreased BIRC5 expression was associated with better clinical outcome (p = 0.004) in AML patients exhibiting CBF mainly due to patients with inv(16) (p = 0.007). This study has shown that BIRC5 expression plays a role in the survival of AML patients, this association is not apparent when we examine CBF patients as a cohort, but when those with inv(16) independently indicating that those patients with inv(16) would provide interesting candidates for immunotherapies that target BIRC5.

Selecting Relevant Genes From Microarray Datasets Using a Random Forest Model

Recent studies have demonstrated microarray expression data can be used to identify gene regulatory pathways. However, one of the major challenges is to utilize the large microarray data (genes and micro-RNAs) to have an efficient computational model. Therefore, there is an urgent need to reduce the dimensionality of these large sets using machine learning methods without compromising the accuracy. This requires an appropriate machine learning algorithm to select the significant features from these large datasets. Therefore, in this study, we use a supervised method based on a Random Forest to identify significant features from three microarray datasets from prenatal nicotine, alcohol, and nicotine and alcohol exposure groups in two different cell types (dopamine and non-dopamine neurons). Our approach was computationally efficient to reduce the dimensionality of extremely large microarray datasets. Furthermore, our results indicated that using only the top 20% of features was sufficient to confirm the genetic pathways previously identified when using all of the features in the model.

A transcriptome-based classifier to determine molecular subtypes in medulloblastoma.

Medulloblastoma is a highly heterogeneous pediatric brain tumor with five molecular subtypes, Sonic Hedgehog TP53-mutant, Sonic Hedgehog TP53-wildtype, WNT, Group 3, and Group 4, defined by the World Health Organization. The current mechanism for classification into these molecular subtypes is through the use of immunostaining, methylation, and/or genetics. We surveyed the literature and identified a number of RNA-Seq and microarray datasets in order to develop, train, test, and validate a robust classifier to identify medulloblastoma molecular subtypes through the use of transcriptomic profiling data. We have developed a GPL-3 licensed R package and a Shiny Application to enable users to quickly and robustly classify medulloblastoma samples using transcriptomic data. The classifier utilizes a large composite microarray dataset (15 individual datasets), an individual microarray study, and an RNA-Seq dataset, using gene ratios instead of gene expression measures as features for the model. Discriminating features were identified using the limma R package and samples were classified using an unweighted mean of normalized scores. We utilized two training datasets and applied the classifier in 15 separate datasets. We observed a minimum accuracy of 85.71% in the smallest dataset and a maximum of 100% accuracy in four datasets with an overall median accuracy of 97.8% across the 15 datasets, with the majority of misclassification occurring between the heterogeneous Group 3 and Group 4 subtypes. We anticipate this medulloblastoma transcriptomic subtype classifier will be broadly applicable to the cancer research and clinical communities.

Transcriptomic Profiling for the Autophagy Pathway in Colorectal Cancer.

The role of autophagy in colorectal cancer (CRC) pathogenesis appears to be crucial. Autophagy acts both as a tumor suppressor, by removing redundant cellular material, and a tumor-promoting factor, by providing access to components necessary for growth, metabolism, and proliferation. To date, little is known about the expression of genes that play a basal role in the autophagy in CRC. In this study, we aimed to compare the expression levels of 46 genes involved in the autophagy pathway between tumor-adjacent and tumor tissue, employing large RNA sequencing (RNA-seq) and microarray datasets. Additionally, we verified our results using data on 38 CRC cell lines. Gene set enrichment analysis revealed a significant deregulation of autophagy-related gene sets in CRC. The unsupervised clustering of tumors using the mRNA levels of autophagy-related genes revealed the existence of two major clusters: microsatellite instability (MSI)-enriched and -depleted. In cluster 1 (MSI-depleted), ATG9B and LAMP1 genes were the most prominently expressed, whereas cluster 2 (MSI-enriched) was characterized by DRAM1 upregulation. CRC cell lines were also clustered according to MSI-enriched/-depleted subgroups. The moderate deregulation of autophagy-related genes in cancer tissue, as compared to adjacent tissue, suggests a prominent field cancerization or early disruption of autophagy. Genes differentiating these clusters are promising candidates for CRC targeting therapy worthy of further investigation.

Blood transcriptome profiling captures dysregulated pathways and response to treatment in neuroimmunological disease.

Blood transcriptome profiling captures dysregulated pathways and response to treatment in neuroimmunological disease.

Generalized gene co-expression analysis via subspace clustering using low-rank representation

BackgroundGene Co-expression Network Analysis (GCNA) helps identify gene modules with potential biological functions and has become a popular method in bioinformatics and biomedical research. However, most current GCNA algorithms use correlation to build gene co-expression networks and identify modules with highly correlated genes. There is a need to look beyond correlation and identify gene modules using other similarity measures for finding novel biologically meaningful modules.ResultsWe propose a new generalized gene co-expression analysis algorithm via subspace clustering that can identify biologically meaningful gene co-expression modules with genes that are not all highly correlated. We use low-rank representation to construct gene co-expression networks and local maximal quasi-clique merger to identify gene co-expression modules. We applied our method on three large microarray datasets and a single-cell RNA sequencing dataset. We demonstrate that our method can identify gene modules with different biological functions than current GCNA methods and find gene modules with prognostic values.ConclusionsThe presented method takes advantage of subspace clustering to generate gene co-expression networks rather than using correlation as the similarity measure between genes. Our generalized GCNA method can provide new insights from gene expression datasets and serve as a complement to current GCNA algorithms.

Epigenetic regulation of placental gene expression in transcriptional subtypes of preeclampsia

BackgroundPreeclampsia (PE) is a heterogeneous, hypertensive disorder of pregnancy, with no robust biomarkers or effective treatments. We hypothesized that this heterogeneity is due to the existence of multiple subtypes of PE and, in support of this hypothesis, we recently identified five clusters of placentas within a large gene expression microarray dataset (N = 330), of which four (clusters 1, 2, 3, and 5) contained a substantial number of PE samples. However, while transcriptional analysis of placentas can subtype patients, we propose that the addition of epigenetic information could discern gene regulatory mechanisms behind the distinct PE pathologies, as well as identify clinically useful potential biomarkers.ResultsWe subjected 48 of our samples from transcriptional clusters 1, 2, 3, and 5 to Infinium HumanMethylation450 arrays. Samples belonging to transcriptional clusters 1–3 still showed visible relationships to each other by methylation, but cluster 5, with known chromosomal abnormalities, no longer formed a cohesive group. Within transcriptional clusters 2 and 3, controlling for fetal sex and gestational age in the identification of differentially methylated sites, compared to the healthier cluster 1, dramatically reduced the number of significant sites, but increased the percentage that demonstrated a strong linear correlation with gene expression (from 5% and 2% to 9% and 8%, respectively). Locations exhibiting a positive relationship between methylation and gene expression were most frequently found in CpG open sea enhancer regions within the gene body, while those with a significant negative correlation were often annotated to the promoter in a CpG shore region. Integrated transcriptome and epigenome analysis revealed modifications in TGF-beta signaling, cell adhesion, oxidative phosphorylation, and metabolism pathways in cluster 2 placentas, and aberrations in antigen presentation, allograft rejection, and cytokine-cytokine receptor interaction in cluster 3 samples.ConclusionsOverall, we have established DNA methylation alterations underlying a portion of the transcriptional development of “canonical” PE in cluster 2 and “immunological” PE in cluster 3. However, a significant number of the observed methylation changes were not associated with corresponding changes in gene expression, and vice versa, indicating that alternate methods of gene regulation will need to be explored to fully comprehend these PE subtypes.

Fuzzy Expert System based on a Novel Hybrid Stem Cell (HSC) Algorithm for Classification of Micro Array Data.

In the growing scenario, microarray data is extensively used since it provides a more comprehensive understanding of genetic variants among diseases. As the gene expression samples have high dimensionality it becomes tedious to analyze the samples manually. Hence an automated system is needed to analyze these samples. The fuzzy expert system offers a clear classification when compared to the machine learning and statistical methodologies. In fuzzy classification, knowledge acquisition would be a major concern. Despite several existing approaches for knowledge acquisition much effort is necessary to enhance the learning process. This paper proposes an innovative Hybrid Stem Cell (HSC) algorithm that utilizes Ant Colony optimization and Stem Cell algorithm for designing fuzzy classification system to extract the informative rules to form the membership functions from the microarray dataset. The HSC algorithm uses a novel Adaptive Stem Cell Optimization (ASCO) to improve the points of membership function and Ant Colony Optimization to produce the near optimum rule set. In order to extract the most informative genes from the large microarray dataset a method called Mutual Information is used. The performance results of the proposed technique evaluated using the five microarray datasets are simulated. These results prove that the proposed Hybrid Stem Cell (HSC) algorithm produces a precise fuzzy system than the existing methodologies.

Comparison of statistical methods for subnetwork detection in the integration of gene expression and protein interaction network

BackgroundWith the advancement of high-throughput technologies and enrichment of popular public databases, more and more research focuses of bioinformatics research have been on computational integration of network and gene expression profiles for extracting context-dependent active subnetworks. Many methods for subnetwork searching have been developed. Scoring and searching algorithms present a range of computational considerations and implementations. The primary goal of present study is to comprehensively evaluate the performance of different subnetwork detection methods. Eleven popular methods were selected for comprehensive comparison.ResultsFirst, taking into account the dependence of genes given a protein-protein interaction (PPI) network, we simulated microarray gene expression data under case and control conditions. Then each method was applied to the simulated data for subnetwork identification. Second, a large microarray data set of prostate cancer was used to assess the practical performance of each method. Using both simulation studies and a real data application, we evaluated the performance of different methods in terms of recall and precision.ConclusionsjActiveModules, PinnacleZ and WMAXC performed well in identifying subnetwork with relative high precision and recall. BioNet performed very well only in precision. As none of methods outperformed other methods overall, users should choose an appropriate method based on the purposes of their studies.

Dynamic Genetic Algorithm-Based Feature Selection and Incomplete Value Imputation for Microarray Classification

Large microarray datasets usually contain many features with missing values. Inferences made from such incomplete datasets may be biased. To address this issue, we propose a novel preprocessing method called dynamic genetic algorithm-based feature selection with missing value imputation. The significant features are first identified using dynamic genetic algorithm-based feature selection and then the missing values are imputed using dynamic Bayesian genetic algorithm. The resulting complete microarray datasets with reduced features are used for classification, which results in better accuracy than the existing methods in eight microarray datasets.

Partial correlation screening for estimating large precision matrices, with applications to classification

We propose Partial Correlation Screening (PCS) as a new row-by-row approach to estimating a large precision matrix $\Omega$. To estimate the $i$-th row of $\Omega$, $1 \leq i \leq p$, PCS uses a Screen step and a Clean step. In the Screen step, PCS recruits a (small) subset of indices using a stage-wise algorithm, where in each stage, the algorithm updates the set of recruited indices by adding the index $j$ that has the largest (in magnitude) empirical partial correlation with $i$. In the Clean step, PCS re-investigates all recruited indices and use them to reconstruct the $i$-th row of $\Omega$. PCS is computationally efficient and modest in memory use: to estimate a row of $\Omega$, it only needs a few rows (determined sequentially) of the empirical covariance matrix. This enables PCS to execute the estimation of a large precision matrix (e.g., $p=10K$) in a few minutes, and open doors to estimating much larger precision matrices. We use PCS for classification. Higher Criticism Thresholding (HCT) is a recent classifier that enjoys optimality, but to exploit its full potential in practice, one needs a good estimate of the precision matrix $\Omega$. Combining HCT with any approach to estimating $\Omega$ gives a new classifier: examples include HCT-PCS and HCT-glasso. We have applied HCT-PCS to two large microarray data sets ($p = 8K$ and $10K$) for classification, where it not only significantly outperforms HCT-glasso, but also is competitive to the Support Vector Machine (SVM) and Random Forest (RF). The results suggest that PCS gives more useful estimates of $\Omega$ than the glasso. We set up a general theoretical framework and show that in a broad context, PCS fully recovers the support of $\Omega$ and HCT-PCS yields optimal classification behavior. Our proofs shed interesting light on the behavior of stage-wise procedures.

Genetic Networks in Mouse Retinal Ganglion Cells.

Retinal ganglion cells (RGCs) are the output neuron of the eye, transmitting visual information from the retina through the optic nerve to the brain. The importance of RGCs for vision is demonstrated in blinding diseases where RGCs are lost, such as in glaucoma or after optic nerve injury. In the present study, we hypothesize that normal RGC function is transcriptionally regulated. To test our hypothesis, we examine large retinal expression microarray datasets from recombinant inbred mouse strains in GeneNetwork and define transcriptional networks of RGCs and their subtypes. Two major and functionally distinct transcriptional networks centering around Thy1 and Tubb3 (Class III beta-tubulin) were identified. Each network is independently regulated and modulated by unique genomic loci. Meta-analysis of publically available data confirms that RGC subtypes are differentially susceptible to death, with alpha-RGCs and intrinsically photosensitive RGCs (ipRGCs) being less sensitive to cell death than other RGC subtypes in a mouse model of glaucoma.

Microarray Data Processing Techniques for Genome-Scale Network Inference from Large Public Repositories.

Pre-processing of microarray data is a well-studied problem. Furthermore, all popular platforms come with their own recommended best practices for differential analysis of genes. However, for genome-scale network inference using microarray data collected from large public repositories, these methods filter out a considerable number of genes. This is primarily due to the effects of aggregating a diverse array of experiments with different technical and biological scenarios. Here we introduce a pre-processing pipeline suitable for inferring genome-scale gene networks from large microarray datasets. We show that partitioning of the available microarray datasets according to biological relevance into tissue- and process-specific categories significantly extends the limits of downstream network construction. We demonstrate the effectiveness of our pre-processing pipeline by inferring genome-scale networks for the model plant Arabidopsis thaliana using two different construction methods and a collection of 11,760 Affymetrix ATH1 microarray chips. Our pre-processing pipeline and the datasets used in this paper are made available at http://alurulab.cc.gatech.edu/microarray-pp.

A COMPARATIVE STUDY ON GENE SELECTION METHODS FOR TISSUES CLASSIFICATION ON LARGE SCALE GENE EXPRESSION DATA

Deoxyribonucleic acid (DNA) microarray technology is the recent invention that provided colossal opportunities to measure a large scale of gene expressions simultaneously. However, interpreting large scale of gene expression data remain a challenging issue due to their innate nature of “high dimensional low sample size”. Microarray data mainly involved thousands of genes, n in a very small size sample, p which complicates the data analysis process. For such a reason, feature selection methods also known as gene selection methods have become apparently need to select significant genes that present the maximum discriminative power between cancerous and normal tissues. Feature selection methods can be structured into three basic factions; a) filter methods; b) wrapper methods and c) embedded methods. Among these methods, filter gene selection methods provide easy way to calculate the informative genes and can simplify reduce the large scale microarray datasets. Although filter based gene selection techniques have been commonly used in analyzing microarray dataset, these techniques have been tested separately in different studies. Therefore, this study aims to investigate and compare the effectiveness of these four popular filter gene selection methods namely Signal-to-Noise ratio (SNR), Fisher Criterion (FC), Information Gain (IG) and t-Test in selecting informative genes that can distinguish cancer and normal tissues. In this experiment, common classifiers, Support Vector Machine (SVM) is used to train the selected genes. These gene selection methods are tested on three large scales of gene expression datasets, namely breast cancer dataset, colon dataset, and lung dataset. This study has discovered that IG and SNR are more suitable to be used with SVM. Furthermore, this study has shown SVM performance remained moderately unaffected unless a very small size of genes was selected.

TIGER: an evolutionary search for Top Inter-GEne Relations

Relative Expression Analysis RXA plays an important role in biomarker discovery and microarray data classification. It focuses on ordering relationships between the expression of small sets of genes rather than their raw values. Most of the RXA algorithms are preceded by feature selection as analysing all possible subsets of genes is computationally infeasible. In this paper, we propose an efficient solution that unifies major variants of RXA algorithms and is capable of searching top inter-gene relations even in large microarray datasets. A specialised evolutionary algorithm that incorporates and exploits knowledge about RXA into the evolutionary search allows exploring solution space with all available genes. By embedding information about the genes' discriminative power we managed to speed up the evolutionary process and to search for complex interactions between genes. Experimental validation shows that the proposed solution outperforms popular RXA algorithms and has considerable potential for discovering new relationships between the genes.

DECODE: an integrated differential co-expression and differential expression analysis of gene expression data.

BackgroundBoth differential expression (DE) and differential co-expression (DC) analyses are appreciated as useful tools in understanding gene regulation related to complex diseases. The performance of integrating DE and DC, however, remains unexplored.ResultsIn this study, we proposed a novel analytical approach called DECODE (Differential Co-expression and Differential Expression) to integrate DC and DE analyses of gene expression data. DECODE allows one to study the combined features of DC and DE of each transcript between two conditions. By incorporating information of the dependency between DC and DE variables, two optimal thresholds for defining substantial change in expression and co-expression are systematically defined for each gene based on chi-square maximization. By using these thresholds, genes can be categorized into four groups with either high or low DC and DE characteristics. In this study, DECODE was applied to a large breast cancer microarray data set consisted of two thousand tumor samples. By identifying genes with high DE and high DC, we demonstrated that DECODE could improve the detection of some functional gene sets such as those related to immune system, metastasis, lipid and glucose metabolism. Further investigation on the identified genes and the associated functional pathways would provide an additional level of understanding of complex disease mechanism.ConclusionsBy complementing the recent DC and the traditional DE analyses, DECODE is a valuable methodology for investigating biological functions of genes exhibiting disease-associated DE and DC combined characteristics, which may not be easily revealed through DC or DE approach alone.DECODE is available at the Comprehensive R Archive Network (CRAN): http://cran.r-project.org/web/packages/decode/index.html.Electronic supplementary materialThe online version of this article (doi:10.1186/s12859-015-0582-4) contains supplementary material, which is available to authorized users.

Rough Based Symmetrical Clustering for Gene Expression Profile Analysis.

Identification of coexpressed genes is the central goal in microarray gene expression data analysis. Point symmetry-based clustering is an important unsupervised learning technique for recognizing symmetrical convex or non-convex shaped clusters. To enable fast automatic clustering of large microarray data, in this article, a distributed time-efficient scalable parallel rough set based hybrid approach for point symmetry-based clustering algorithm has been proposed. A natural basis for analyzing gene expression data using the symmetry-based algorithm, is to group together genes with similar symmetrical patterns of expression. Rough-set theory helps in faster convergence and initial automatic optimal classification, thereby solving the problem of unknown knowledge of number of clusters in microarray data. This new parallel implementation with K-means algorithm also satisfies the linear speedup in timing on large microarray datasets. This proposed algorithm is compared with another parallel symmetry-based K-means and parallel version of existing K-means over four artificial and benchmark microarray datasets. We also have experimented over three skewed cancer gene expression datasets. The statistical analysis are also performed to establish the significance of this new implementation. The biological relevance of the clustering solutions are also analyzed.

Immune gene signatures and tumor intrinsic markers delineate novel immunogenic subtypes of breast cancer

The abundance and functional orientation of tumor-infiltrating effector cells has long been observed to predict for reduced incidence of clinical metastasis and cancer-specific death. Using bioinformatics to mine large breast tumor microarray datasets, we and others have identified prognostic immune gene signatures, or metagenes. Robust evidence indicates that these metagenes are: 1) positively correlated with distant metastasis-free survival (DMFS) of patients, 2) comprised of genes that regulate immune cell-specific biology, and 3) reflective of the relative abundance of discernible populations of tumor infiltrating leukocytes. In recent work we have leveraged the statistical associations between the immune metagenes and the DMFS of breast cancer patients to explore the underlying phenotypes that differ in their ability to potentiate long-term, immune-mediated tumor rejection. Using a tumor classification model that combines the prognostic attributes of three distinct immune metagenes, termed the B/P, T/NK and M/D metagenes, we have identified molecular subtypes of breast cancer that either permit or prohibit prognostication by the immune metagenes. On this basis, we have delineated the phenotypic attributes of breast cancer that distinguish two novel immunogenic tumor subtypes, which we have defined as: immune benefit-enabled (IBE) and immune benefit-disabled (IBD). Phenotypically, IBE tumors comprise of Basal-like tumors and highly-proliferative HER2-Enriched and Luminal-B subtypes, while IBD tumors comprise of Claudin-Low, Luminal-A, and low-proliferative HER2-Enriched and Luminal-B tumors. Prognostically, IBE tumors (n = 666) can be stratified by the immune metagene model into prognostic subgroups with high statistical significance (P<0.0001,log-rank test), while IBD tumors cannot (n = 1005, P = 0.3) consistent with the capacity for an innate anti-tumor immunity against IBE tumors, but not IBD tumors, that guards against distant metastasis. Furthermore, these observations were independent of adjuvant treatment, and may owe to differential activation of immunomodulatory pathways. Network analysis revealed that IBE/IBD differentially-expressed genes (q<0.01) underlie highly-significant pathway activation scores for TGF-beta signaling in IBD (p < 0.0001), and Interferon-gamma signaling in IBE (p < 0.0001). Furthermore, 15 of 19 genes comprising the previously described Immunologic Constant of Rejection (Marincola and colleagues) were significantly overexpressed in IBE tumors (P-value range: 0.05-3.5E-14). Thus, we conclude that breast tumors can be dichotomized into two subtypes fundamentally distinct with respect to their potential for metastasis-protective immune responsiveness. These findings indicate new contexts for studying anti-tumor immunity and oncogenic mechanisms of immunosuppression in breast cancer. Whether IBE and IBD subtypes represent clinically-relevant contexts for assessing patient prognosis or evaluating the efficacy of immunotherapeutic treatments warrants further investigation (See Figure ​Figure11). Figure 1 (A-D) The immune metagenes are prognostic of IBE but no IBD breast cancer. (A) Heatmaps of metagene expression levels (rows) across 1,954 tumors (columns). Key shows color scale of mean centered, log2-transformed gene signal intensities. For each metagene, ...