Sabri Boughorbel

Optimal classifier for imbalanced data using Matthews Correlation Coefficient metric

Data imbalance is frequently encountered in biomedical applications. Resampling techniques can be used in binary classification to tackle this issue. However such solutions are not desired when the number of samples in the small class is limited. Moreover the use of inadequate performance metrics, such as accuracy, lead to poor generalization results because the classifiers tend to predict the largest size class. One of the good approaches to deal with this issue is to optimize performance metrics that are designed to handle data imbalance. Matthews Correlation Coefficient (MCC) is widely used in Bioinformatics as a performance metric. We are interested in developing a new classifier based on the MCC metric to handle imbalanced data. We derive an optimal Bayes classifier for the MCC metric using an approach based on Frechet derivative. We show that the proposed algorithm has the nice theoretical property of consistency. Using simulated data, we verify the correctness of our optimality result by searching in the space of all possible binary classifiers. The proposed classifier is evaluated on 64 datasets from a wide range data imbalance. We compare both classification performance and CPU efficiency for three classifiers: 1) the proposed algorithm (MCC-classifier), the Bayes classifier with a default threshold (MCC-base) and imbalanced SVM (SVM-imba). The experimental evaluation shows that MCC-classifier has a close performance to SVM-imba while being simpler and more efficient.

Model Comparison for Breast Cancer Prognosis Based on Clinical Data.

We compared the performance of several prediction techniques for breast cancer prognosis, based on AU-ROC performance (Area Under ROC) for different prognosis periods. The analyzed dataset contained 1,981 patients and from an initial 25 variables, the 11 most common clinical predictors were retained. We compared eight models from a wide spectrum of predictive models, namely; Generalized Linear Model (GLM), GLM-Net, Partial Least Square (PLS), Support Vector Machines (SVM), Random Forests (RF), Neural Networks, k-Nearest Neighbors (k-NN) and Boosted Trees. In order to compare these models, paired t-test was applied on the model performance differences obtained from data resampling. Random Forests, Boosted Trees, Partial Least Square and GLMNet have superior overall performance, however they are only slightly higher than the other models. The comparative analysis also allowed us to define a relative variable importance as the average of variable importance from the different models. Two sets of variables are identified from this analysis. The first includes number of positive lymph nodes, tumor size, cancer grade and estrogen receptor, all has an important influence on model predictability. The second set incudes variables related to histological parameters and treatment types. The short term vs long term contribution of the clinical variables are also analyzed from the comparative models. From the various cancer treatment plans, the combination of Chemo/Radio therapy leads to the largest impact on cancer prognosis.

IRF4 haploinsufficiency in a family with Whipple’s disease

Most humans are exposed to Tropheryma whipplei (Tw). Whipple’s disease (WD) strikes only a small minority of individuals infected with Tw (<0.01%), whereas asymptomatic chronic carriage is more common (<25%). We studied a multiplex kindred, containing four WD patients and five healthy Tw chronic carriers. We hypothesized that WD displays autosomal dominant (AD) inheritance, with age-dependent incomplete penetrance. We identified a single very rare non-synonymous mutation in the four patients: the private R98W variant of IRF4, a transcription factor involved in immunity. The five Tw carriers were younger, and also heterozygous for R98W. We found that R98W was loss-of-function, modified the transcriptome of heterozygous leukocytes following Tw stimulation, and was not dominant-negative. We also found that only six of the other 153 known non-synonymous IRF4 variants were loss-of-function. Finally, we found that IRF4 had evolved under purifying selection. AD IRF4 deficiency can underlie WD by haploinsufficiency, with age-dependent incomplete penetrance.

Inherited human IRAK-1 deficiency selectively impairs TLR signaling in fibroblasts

Significance We report the discovery of complete human interleukin-1 receptor (IL-1R)-associated kinase 1 (IRAK-1) deficiency resulting from a de novo Xq28 microdeletion encompassing MECP2 and IRAK1 in a boy. Like many boys with MECP2 defects, this patient died very early. IRAK-1 is a component of the Toll-like receptor (TLR)/IL-1R (TIR) signaling pathway. Unlike patients with autosomal-recessive complete deficiency of MyD88 or IRAK-4, two other components of the TIR pathway, this patient presented no invasive bacterial infections. We analyzed the impact of human IRAK-1 deficiency in fibroblasts and leukocytes. The role of IRAK-1 in signaling downstream from IL-1R and TLRs differed according to cell type. These findings reveal similarities and differences in the role of IRAK-1 in the TLR and IL-1R pathways between mice and humans. Most members of the Toll-like receptor (TLR) and interleukin-1 receptor (IL-1R) families transduce signals via a canonical pathway involving the MyD88 adapter and the interleukin-1 receptor-associated kinase (IRAK) complex. This complex contains four molecules, including at least two (IRAK-1 and IRAK-4) active kinases. In mice and humans, deficiencies of IRAK-4 or MyD88 abolish most TLR (except for TLR3 and some TLR4) and IL-1R signaling in both leukocytes and fibroblasts. TLR and IL-1R responses are weak but not abolished in mice lacking IRAK-1, whereas the role of IRAK-1 in humans remains unclear. We describe here a boy with X-linked MECP2 deficiency-related syndrome due to a large de novo Xq28 chromosomal deletion encompassing both MECP2 and IRAK1. Like many boys with MECP2 null mutations, this child died very early, at the age of 7 mo. Unlike most IRAK-4– or MyD88-deficient patients, he did not suffer from invasive bacterial diseases during his short life. The IRAK-1 protein was completely absent from the patient’s fibroblasts, which responded very poorly to all TLR2/6 (PAM2CSK4, LTA, FSL-1), TLR1/2 (PAM3CSK4), and TLR4 (LPS, MPLA) agonists tested but had almost unimpaired responses to IL-1β. By contrast, the patient’s peripheral blood mononuclear cells responded normally to all TLR1/2, TLR2/6, TLR4, TLR7, and TLR8 (R848) agonists tested, and to IL-1β. The death of this child precluded long-term evaluations of the clinical consequences of inherited IRAK-1 deficiency. However, these findings suggest that human IRAK-1 is essential downstream from TLRs but not IL-1Rs in fibroblasts, whereas it plays a redundant role downstream from both TLRs and IL-1Rs in leukocytes.

A curated compendium of monocyte transcriptome datasets of relevance to human monocyte immunobiology research

Systems-scale profiling approaches have become widely used in translational research settings. The resulting accumulation of large-scale datasets in public repositories represents a critical opportunity to promote insight and foster knowledge discovery. However, resources that can serve as an interface between biomedical researchers and such vast and heterogeneous dataset collections are needed in order to fulfill this potential. Recently, we have developed an interactive data browsing and visualization web application, the Gene Expression Browser (GXB). This tool can be used to overlay deep molecular phenotyping data with rich contextual information about analytes, samples and studies along with ancillary clinical or immunological profiling data. In this note, we describe a curated compendium of 93 public datasets generated in the context of human monocyte immunological studies, representing a total of 4,516 transcriptome profiles. Datasets were uploaded to an instance of GXB along with study description and sample annotations. Study samples were arranged in different groups. Ranked gene lists were generated based on relevant group comparisons. This resource is publicly available online at http://monocyte.gxbsidra.org/dm3/landing.gsp.

A curated transcriptome dataset collection to investigate the development and differentiation of the human placenta and its associated pathologies

Compendia of large-scale datasets made available in public repositories provide a precious opportunity to discover new biomedical phenomena and to fill gaps in our current knowledge. In order to foster novel insights it is necessary to ensure that these data are made readily accessible to research investigators in an interpretable format. Here we make a curated, public, collection of transcriptome datasets relevant to human placenta biology available for further analysis and interpretation via an interactive data browsing interface. We identified and retrieved a total of 24 datasets encompassing 759 transcriptome profiles associated with the development of the human placenta and associated pathologies from the NCBI Gene Expression Omnibus (GEO) and present them in a custom web-based application designed for interactive query and visualization of integrated large-scale datasets ( http://placentalendocrinology.gxbsidra.org/dm3/landing.gsp). We also performed quality control checks using relevant biological markers. Multiple sample groupings and rank lists were subsequently created to facilitate data query and interpretation. Via this interface, users can create web-links to customized graphical views which may be inserted into manuscripts for further dissemination, or e-mailed to collaborators for discussion. The tool also enables users to browse a single gene across different projects, providing a mechanism for developing new perspectives on the role of a molecule of interest across multiple biological states. The dataset collection we created here is available at: http://placentalendocrinology.gxbsidra.org/dm3.

A collection of annotated and harmonized human breast cancer transcriptome datasets, including immunologic classification

The increased application of high-throughput approaches in translational research has expanded the number of publicly available data repositories. Gathering additional valuable information contained in the datasets represents a crucial opportunity in the biomedical field. To facilitate and stimulate utilization of these datasets, we have recently developed an interactive data browsing and visualization web application, the Gene Expression Browser (GXB). In this note, we describe a curated compendium of 13 public datasets on human breast cancer, representing a total of 2142 transcriptome profiles. We classified the samples according to different immune based classification systems and integrated this information into the datasets. Annotated and harmonized datasets were uploaded to GXB. Study samples were categorized in different groups based on their immunologic tumor response profiles, intrinsic molecular subtypes and multiple clinical parameters. Ranked gene lists were generated based on relevant group comparisons. In this data note, we demonstrate the utility of GXB to evaluate the expression of a gene of interest, find differential gene expression between groups and investigate potential associations between variables with a specific focus on immunologic classification in breast cancer. This interactive resource is publicly available online at: http://breastcancer.gxbsidra.org/dm3/geneBrowser/list.

A curated transcriptomic dataset collection relevant to embryonic development associated with in vitro fertilization in healthy individuals and patients with polycystic ovary syndrome

The collection of large-scale datasets available in public repositories is rapidly growing and providing opportunities to identify and fill gaps in different fields of biomedical research. However, users of these datasets should be able to selectively browse datasets related to their field of interest. Here we made available a collection of transcriptome datasets related to human follicular cells from normal individuals or patients with polycystic ovary syndrome, in the process of their development, during in vitro fertilization. After RNA-seq dataset exclusion and careful selection based on study description and sample information, 12 datasets, encompassing a total of 85 unique transcriptome profiles, were identified in NCBI Gene Expression Omnibus and uploaded to the Gene Expression Browser (GXB), a web application specifically designed for interactive query and visualization of integrated large-scale data. Once annotated in GXB, multiple sample grouping has been made in order to create rank lists to allow easy data interpretation and comparison. The GXB tool also allows the users to browse a single gene across multiple projects to evaluate its expression profiles in multiple biological systems/conditions in a web-based customized graphical views. The curated dataset is accessible at the following link: http://ivf.gxbsidra.org/dm3/landing.gsp.

A compendium of monocyte transcriptome datasets to foster biomedical knowledge discovery

Systems-scale profiling approaches have become widely used in translational research settings. The resulting accumulation of large-scale datasets in public repositories represents a critical opportunity to promote insight and foster knowledge discovery. However, resources that can serve as an interface between biomedical researchers and such vast and heterogeneous dataset collections are needed in order to fulfill this potential. Recently, we have developed an interactive data browsing and visualization web application, the Gene Expression Browser (GXB). This tool can be used to overlay deep molecular phenotyping data with rich contextual information about analytes, samples and studies along with ancillary clinical or immunological profiling data. In this note, we describe a curated compendium of 93 public datasets generated in the context of human monocyte immunological studies, representing a total of 4,516 transcriptome profiles. Datasets were uploaded to an instance of GXB along with study description and sample annotations. Study samples were arranged in different groups. Ranked gene lists were generated based on relevant group comparisons. This resource is publicly available online at http://monocyte.gxbsidra.org/dm3/landing.gsp.

A curated transcriptome dataset collection to investigate the functional programming of human hematopoietic cells in early life

Compendia of large-scale datasets made available in public repositories provide an opportunity to identify and fill gaps in biomedical knowledge. But first, these data need to be made readily accessible to research investigators for interpretation. Here we make available a collection of transcriptome datasets to investigate the functional programming of human hematopoietic cells in early life. Thirty two datasets were retrieved from the NCBI Gene Expression Omnibus (GEO) and loaded in a custom web application called the Gene Expression Browser (GXB), which was designed for interactive query and visualization of integrated large-scale data. Quality control checks were performed. Multiple sample groupings and gene rank lists were created allowing users to reveal age-related differences in transcriptome profiles, changes in the gene expression of neonatal hematopoietic cells to a variety of immune stimulators and modulators, as well as during cell differentiation. Available demographic, clinical, and cell phenotypic information can be overlaid with the gene expression data and used to sort samples. Web links to customized graphical views can be generated and subsequently inserted in manuscripts to report novel findings. GXB also enables browsing of a single gene across projects, thereby providing new perspectives on age- and developmental stage-specific expression of a given gene across the human hematopoietic system. This dataset collection is available at: http://developmentalimmunology.gxbsidra.org/dm3/geneBrowser/list.