OBJECTIVE: Postpartum hemorrhage (PPH) remains a leading cause of preventable maternal mortality in the United States. We sought to develop a novel risk assessment tool and compare its accuracy to tools used in current practice. MATERIALS AND METHODS: We used a PPH digital phenotype that we developed and validated previously to identify 6639 PPH deliveries from our delivery cohort (N = 70 948). Using a vast array of known and potential risk factors extracted from electronic medical records available prior to delivery, we trained a gradient boosting model in a subset of our cohort. In a held-out test sample, we compared performance of our model with 3 clinical risk-assessment tools and 1 previously published model. RESULTS: Our 24-feature model achieved an area under the receiver-operating characteristic curve (AUROC) of 0.71 (95% confidence interval [CI], 0.69-0.72), higher than all other tools (research-based AUROC, 0.67 [95% CI, 0.66-0.69]; clinical AUROCs, 0.55 [95% CI, 0.54-0.56] to 0.61 [95% CI, 0.59-0.62]). Five features were novel, including red blood cell indices and infection markers measured upon admission. Additionally, we identified inflection points for vital signs and labs where risk rose substantially. Most notably, patients with median intrapartum systolic blood pressure above 132 mm Hg had an 11% (95% CI, 8%-13%) median increase in relative risk for PPH. CONCLUSIONS: We developed a novel approach for predicting PPH and identified clinical feature thresholds that can guide intrapartum monitoring for PPH risk. These results suggest that our model is an excellent candidate for prospective evaluation and could ultimately reduce PPH morbidity and mortality through early detection and prevention.

OBJECTIVE: We aimed to establish a comprehensive digital phenotype for postpartum hemorrhage (PPH). Current guidelines rely primarily on estimates of blood loss, which can be inaccurate and biased and ignore complementary information readily available in electronic medical records (EMR). Inaccurate and incomplete phenotyping contributes to ongoing challenges in tracking PPH outcomes, developing more accurate risk assessments, and identifying novel interventions. MATERIALS AND METHODS: We constructed a cohort of 71 944 deliveries from the Mount Sinai Health System. Estimates of postpartum blood loss, shifts in hematocrit, administration of uterotonics, surgical interventions, and diagnostic codes were combined to identify PPH, retrospectively. Clinical features were extracted from EMRs and mapped to common data models for maximum interoperability across hospitals. Blinded chart review was done by a physician on a subset of PPH and non-PPH patients and performance was compared to alternate PPH phenotypes. PPH was defined as clinical diagnosis of postpartum hemorrhage documented in the patient's chart upon chart review. RESULTS: We identified 6639 PPH deliveries (9% prevalence) using our phenotype-more than 3 times as many as using blood loss alone (N = 1,747), supporting the need to incorporate other diagnostic and intervention data. Chart review revealed our phenotype had 89% accuracy and an F1-score of 0.92. Alternate phenotypes were less accurate, including a common blood loss-based definition (67%) and a previously published digital phenotype (74%). CONCLUSION: We have developed a scalable, accurate, and valid digital phenotype that may be of significant use for tracking outcomes and ongoing clinical research to deliver better preventative interventions for PPH.

Insulin resistance (IR) precedes the development of type 2 diabetes (T2D) and increases cardiovascular disease risk. Although genome wide association studies (GWAS) have uncovered new loci associated with T2D, their contribution to explain the mechanisms leading to decreased insulin sensitivity has been very limited. Thus, new approaches are necessary to explore the genetic architecture of insulin resistance. To that end, we generated an iPSC library across the spectrum of insulin sensitivity in humans. RNA-seq based analysis of 310 induced pluripotent stem cell (iPSC) clones derived from 100 individuals allowed us to identify differentially expressed genes between insulin resistant and sensitive iPSC lines. Analysis of the co-expression architecture uncovered several insulin sensitivity-relevant gene sub-networks, and predictive network modeling identified a set of key driver genes that regulate these co-expression modules. Functional validation in human adipocytes and skeletal muscle cells (SKMCs) confirmed the relevance of the key driver candidate genes for insulin responsiveness.

MOTIVATION: Non-coding rare variants (RVs) may contribute to Mendelian disorders but have been challenging to study due to small sample sizes, genetic heterogeneity and uncertainty about relevant non-coding features. Previous studies identified RVs associated with expression outliers, but varying outlier definitions were employed and no comprehensive open-source software was developed. RESULTS: We developed Outlier-RV Enrichment (ORE) to identify biologically-meaningful non-coding RVs. We implemented ORE combining whole-genome sequencing and cardiac RNAseq from congenital heart defect patients from the Pediatric Cardiac Genomics Consortium and deceased adults from Genotype-Tissue Expression. Use of rank-based outliers maximized sensitivity while a most extreme outlier approach maximized specificity. Rarer variants had stronger associations, suggesting they are under negative selective pressure and providing a basis for investigating their contribution to Mendelian disorders. AVAILABILITY AND IMPLEMENTATION: ORE, source code, and documentation are available at https://pypi.python.org/pypi/ore under the MIT license. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Motivation: The biological relevance of chimeric RNA alignments is now well established. Chimera arising as chromosomal fusions are often drivers of cancer and recently discovered circular RNA (circRNA) are only now being characterized. While software already exists for fusion discovery and quantitation, high false positive rates and high run-times hamper scalable fusion discovery on large datasets. Furthermore, software available for circRNA detection and quantification is limited. Results: Here, we present STAR Chimeric Post (STARChip), a novel software package that processes chimeric alignments from the STAR aligner and produces annotated circRNA and high precision fusions in a rapid, efficient and scalable manner that is appropriate for high dimensional medical omics datasets. Availability and implementation: STARChip is available at https://github.com/LosicLab/STARChip. Supplementary information: Supplementary data are available at Bioinformatics online.

MOTIVATION: Underrepresentation of racial groups represents an important challenge and major gap in phenomics research. Most of the current human phenomics research is based primarily on European populations; hence it is an important challenge to expand it to consider other population groups. One approach is to utilize data from EMR databases that contain patient data from diverse demographics and ancestries. The implications of this racial underrepresentation of data can be profound regarding effects on the healthcare delivery and actionability. To the best of our knowledge, our work is the first attempt to perform comparative, population-scale analyses of disease networks across three different populations, namely Caucasian (EA), African American (AA) and Hispanic/Latino (HL). RESULTS: We compared susceptibility profiles and temporal connectivity patterns for 1988 diseases and 37 282 disease pairs represented in a clinical population of 1 025 573 patients. Accordingly, we revealed appreciable differences in disease susceptibility, temporal patterns, network structure and underlying disease connections between EA, AA and HL populations. We found 2158 significantly comorbid diseases for the EA cohort, 3265 for AA and 672 for HL. We further outlined key disease pair associations unique to each population as well as categorical enrichments of these pairs. Finally, we identified 51 key 'hub' diseases that are the focal points in the race-centric networks and of particular clinical importance. Incorporating race-specific disease comorbidity patterns will produce a more accurate and complete picture of the disease landscape overall and could support more precise understanding of disease relationships and patient management towards improved clinical outcomes. CONTACTS: [email protected] or [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: MicroRNAs (miRNAs) play a key role in regulating tumor progression and metastasis. Identifying key miRNAs, defined by their functional activities, can provide a deeper understanding of biology of miRNAs in cancer. However, miRNA expression level cannot accurately reflect miRNA activity. RESULTS: We developed a computational approach, ActMiR, for identifying active miRNAs and miRNA-mediated regulatory mechanisms. Applying ActMiR to four cancer datasets in The Cancer Genome Atlas (TCGA), we showed that (i) miRNA activity was tumor subtype specific; (ii) genes correlated with inferred miRNA activities were more likely to enrich for miRNA binding motifs; (iii) expression levels of these genes and inferred miRNA activities were more likely to be negatively correlated. For the four cancer types in TCGA we identified 77-229 key miRNAs for each cancer subtype and annotated their biological functions. The miRNA-target pairs, predicted by our ActMiR algorithm but not by correlation of miRNA expression levels, were experimentally validated. The functional activities of key miRNAs were further demonstrated to be associated with clinical outcomes for other cancer types using independent datasets. For ER(-)/HER2(-) breast cancers, we identified activities of key miRNAs let-7d and miR-18a as potential prognostic markers and validated them in two independent ER(-)/HER2(-) breast cancer datasets. Our work provides a novel scheme to facilitate our understanding of miRNA. In summary, inferred activity of key miRNA provided a functional link to its mediated regulatory network, and can be used to robustly predict patient's survival. AVAILABILITY AND IMPLEMENTATION: the software is freely available at http://research.mssm.edu/integrative-network-biology/Software.html. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: As large-scale studies of gene expression with multiple sources of biological and technical variation become widely adopted, characterizing these drivers of variation becomes essential to understanding disease biology and regulatory genetics. RESULTS: We describe a statistical and visualization framework, variancePartition, to prioritize drivers of variation based on a genome-wide summary, and identify genes that deviate from the genome-wide trend. Using a linear mixed model, variancePartition quantifies variation in each expression trait attributable to differences in disease status, sex, cell or tissue type, ancestry, genetic background, experimental stimulus, or technical variables. Analysis of four large-scale transcriptome profiling datasets illustrates that variancePartition recovers striking patterns of biological and technical variation that are reproducible across multiple datasets. CONCLUSIONS: Our open source software, variancePartition, enables rapid interpretation of complex gene expression studies as well as other high-throughput genomics assays. variancePartition is available from Bioconductor: http://bioconductor.org/packages/variancePartition .

Throughout the biomedical and life sciences research community, advanced integrative biology algorithms are employed to integrate large scale data across many different high-dimensional datatypes to construct predictive network models of disease. The causal inference approaches we employ for this purpose well complement the types of natural artificial intelligence/machine learning approaches that have become nearly standard in the life and biomedical sciences for building classifiers for a range of problems, from disease classification and subtype stratification, to the identification of responders and non-responders for a given treatment strategy. By building a causal network model that spans multiple scales (from the molecular to the cellular, to the tissue/organ, to the organism and community) we can understand the flow of information and how best to modulate that flow to improve human wellbeing, whether better diagnosing and treating disease or improving overall health(1-4). More specifically, we have constructed predictive network models for Alzheimer's disease, along with other common human diseases such as obesity, diabetes, heart disease, and inflammatory bowel disease, and cancer, and demonstrated a causal network common across all of these diseases(3, 5-10). Not only do we demonstrate that our predictive models uncover important mechanisms of disease and mechanistic connections among different diseases, but that they have led to a natural way to prioritize therapeutic points of intervention and provide optimal molecular phenotypes for high throughput screening. Our application of these models in a number of disease areas has led to the identification of novel genes that are causal for disease and that may serve as efficacious points of therapeutic intervention, as well as to personalized treatment strategies that provide a more quantitative and accurate approach to tailoring treatments to specific forms of disease.

UNLABELLED: The linear mixed model is the state-of-the-art method to account for the confounding effects of kinship and population structure in genome-wide association studies (GWAS). Current implementations test the effect of one or more genetic markers while including prespecified covariates such as sex. Here we develop an efficient implementation of the linear mixed model that allows composite hypothesis tests to consider genotype interactions with variables such as other genotypes, environment, sex or ancestry. Our R package, lrgpr, allows interactive model fitting and examination of regression diagnostics to facilitate exploratory data analysis in the context of the linear mixed model. By leveraging parallel and out-of-core computing for datasets too large to fit in main memory, lrgpr is applicable to large GWAS datasets and next-generation sequencing data. AVAILABILITY AND IMPLEMENTATION: lrgpr is an R package available from lrgpr.r-forge.r-project.org.

Errors in sample annotation or labeling often occur in large-scale genetic or genomic studies and are difficult to avoid completely during data generation and management. For integrative genomic studies, it is critical to identify and correct these errors. Different types of genetic and genomic data are inter-connected by cis-regulations. On that basis, we developed a computational approach, Multi-Omics Data Matcher (MODMatcher), to identify and correct sample labeling errors in multiple types of molecular data, which can be used in further integrative analysis. Our results indicate that inspection of sample annotation and labeling error is an indispensable data quality assurance step. Applied to a large lung genomic study, MODMatcher increased statistically significant genetic associations and genomic correlations by more than two-fold. In a simulation study, MODMatcher provided more robust results by using three types of omics data than two types of omics data. We further demonstrate that MODMatcher can be broadly applied to large genomic data sets containing multiple types of omics data, such as The Cancer Genome Atlas (TCGA) data sets.

The development of a personalized approach to medical care is now well recognized as an urgent priority. This approach is particularly important in oncology, where it is well understood that each cancer diagnosis is unique at the molecular level, arising from a particular and specific collection of genetic alterations. Furthermore, taking a personalized approach to oncology may expedite the treatment process, pre-empting therapeutic decisions based on fewer data in favor of treatments targeted to an individual's tumor. This directed course may be key to survival for many patients who are terminal or have failed standard therapies.

Breast cancer is the most common malignancy in women and is responsible for hundreds of thousands of deaths annually. As with most cancers, it is a heterogeneous disease and different breast cancer subtypes are treated differently. Understanding the difference in prognosis for breast cancer based on its molecular and phenotypic features is one avenue for improving treatment by matching the proper treatment with molecular subtypes of the disease. In this work, we employed a competition-based approach to modeling breast cancer prognosis using large datasets containing genomic and clinical information and an online real-time leaderboard program used to speed feedback to the modeling team and to encourage each modeler to work towards achieving a higher ranked submission. We find that machine learning methods combined with molecular features selected based on expert prior knowledge can improve survival predictions compared to current best-in-class methodologies and that ensemble models trained across multiple user submissions systematically outperform individual models within the ensemble. We also find that model scores are highly consistent across multiple independent evaluations. This study serves as the pilot phase of a much larger competition open to the whole research community, with the goal of understanding general strategies for model optimization using clinical and molecular profiling data and providing an objective, transparent system for assessing prognostic models.

DNA modifications such as methylation and DNA damage can play critical regulatory roles in biological systems. Single molecule, real time (SMRT) sequencing technology generates DNA sequences as well as DNA polymerase kinetic information that can be used for the direct detection of DNA modifications. We demonstrate that local sequence context has a strong impact on DNA polymerase kinetics in the neighborhood of the incorporation site during the DNA synthesis reaction, allowing for the possibility of estimating the expected kinetic rate of the enzyme at the incorporation site using kinetic rate information collected from existing SMRT sequencing data (historical data) covering the same local sequence contexts of interest. We develop an Empirical Bayesian hierarchical model for incorporating historical data. Our results show that the model could greatly increase DNA modification detection accuracy, and reduce requirement of control data coverage. For some DNA modifications that have a strong signal, a control sample is not even needed by using historical data as alternative to control. Thus, sequencing costs can be greatly reduced by using the model. We implemented the model in a R package named seqPatch, which is available at https://github.com/zhixingfeng/seqPatch.

MOTIVATION: The identification of condition specific sub-networks from gene expression profiles has important biological applications, ranging from the selection of disease-related biomarkers to the discovery of pathway alterations across different phenotypes. Although many methods exist for extracting these sub-networks, very few existing approaches simultaneously consider both the differential expression of individual genes and the differential correlation of gene pairs, losing potentially valuable information in the data. RESULTS: In this article, we propose a new method, COSINE (COndition SpecIfic sub-NEtwork), which employs a scoring function that jointly measures the condition-specific changes of both 'nodes' (individual genes) and 'edges' (gene-gene co-expression). It uses the genetic algorithm to search for the single optimal sub-network which maximizes the scoring function. We applied COSINE to both simulated datasets with various differential expression patterns, and three real datasets, one prostate cancer dataset, a second one from the across-tissue comparison of morbidly obese patients and the other from the across-population comparison of the HapMap samples. Compared with previous methods, COSINE is more powerful in identifying truly significant sub-networks of appropriate size and meaningful biological relevance. AVAILABILITY: The R code is available as the COSINE package on CRAN: http://cran.r-project.org/web/packages/COSINE/index.html.

Many genome-wide datasets are routinely generated to study different aspects of biological systems, but integrating them to obtain a coherent view of the underlying biology remains a challenge. We propose simultaneous clustering of multiple networks as a framework to integrate large-scale datasets on the interactions among and activities of cellular components. Specifically, we develop an algorithm JointCluster that finds sets of genes that cluster well in multiple networks of interest, such as coexpression networks summarizing correlations among the expression profiles of genes and physical networks describing protein-protein and protein-DNA interactions among genes or gene-products. Our algorithm provides an efficient solution to a well-defined problem of jointly clustering networks, using techniques that permit certain theoretical guarantees on the quality of the detected clustering relative to the optimal clustering. These guarantees coupled with an effective scaling heuristic and the flexibility to handle multiple heterogeneous networks make our method JointCluster an advance over earlier approaches. Simulation results showed JointCluster to be more robust than alternate methods in recovering clusters implanted in networks with high false positive rates. In systematic evaluation of JointCluster and some earlier approaches for combined analysis of the yeast physical network and two gene expression datasets under glucose and ethanol growth conditions, JointCluster discovers clusters that are more consistently enriched for various reference classes capturing different aspects of yeast biology or yield better coverage of the analysed genes. These robust clusters, which are supported across multiple genomic datasets and diverse reference classes, agree with known biology of yeast under these growth conditions, elucidate the genetic control of coordinated transcription, and enable functional predictions for a number of uncharacterized genes.

Genetic interactions occur when a combination of mutations results in a surprising phenotype. These interactions capture functional redundancy, and thus are important for predicting function, dissecting protein complexes into functional pathways, and exploring the mechanistic underpinnings of common human diseases. Synthetic sickness and lethality are the most studied types of genetic interactions in yeast. However, even in yeast, only a small proportion of gene pairs have been tested for genetic interactions due to the large number of possible combinations of gene pairs. To expand the set of known synthetic lethal (SL) interactions, we have devised an integrative, multi-network approach for predicting these interactions that significantly improves upon the existing approaches. First, we defined a large number of features for characterizing the relationships between pairs of genes from various data sources. In particular, these features are independent of the known SL interactions, in contrast to some previous approaches. Using these features, we developed a non-parametric multi-classifier system for predicting SL interactions that enabled the simultaneous use of multiple classification procedures. Several comprehensive experiments demonstrated that the SL-independent features in conjunction with the advanced classification scheme led to an improved performance when compared to the current state of the art method. Using this approach, we derived the first yeast transcription factor genetic interaction network, part of which was well supported by literature. We also used this approach to predict SL interactions between all non-essential gene pairs in yeast (http://sage.fhcrc.org/downloads/downloads/predicted_yeast_genetic_interactions.zip). This integrative approach is expected to be more effective and robust in uncovering new genetic interactions from the tens of millions of unknown gene pairs in yeast and from the hundreds of millions of gene pairs in higher organisms like mouse and human, in which very few genetic interactions have been identified to date.

Studies of the relationship between DNA variation and gene expression variation, often referred to as "expression quantitative trait loci (eQTL) mapping", have been conducted in many species and resulted in many significant findings. Because of the large number of genes and genetic markers in such analyses, it is extremely challenging to discover how a small number of eQTLs interact with each other to affect mRNA expression levels for a set of co-regulated genes. We present a Bayesian method to facilitate the task, in which co-expressed genes mapped to a common set of markers are treated as a module characterized by latent indicator variables. A Markov chain Monte Carlo algorithm is designed to search simultaneously for the module genes and their linked markers. We show by simulations that this method is more powerful for detecting true eQTLs and their target genes than traditional QTL mapping methods. We applied the procedure to a data set consisting of gene expression and genotypes for 112 segregants of S. cerevisiae. Our method identified modules containing genes mapped to previously reported eQTL hot spots, and dissected these large eQTL hot spots into several modules corresponding to possibly different biological functions or primary and secondary responses to regulatory perturbations. In addition, we identified nine modules associated with pairs of eQTLs, of which two have been previously reported. We demonstrated that one of the novel modules containing many daughter-cell expressed genes is regulated by AMN1 and BPH1. In conclusion, the Bayesian partition method which simultaneously considers all traits and all markers is more powerful for detecting both pleiotropic and epistatic effects based on both simulated and empirical data.

Gene expression data generated systematically in a given system over multiple time points provides a source of perturbation that can be leveraged to infer causal relationships among genes explaining network changes. Previously, we showed that food intake has a large impact on blood gene expression patterns and that these responses, either in terms of gene expression level or gene-gene connectivity, are strongly associated with metabolic diseases. In this study, we explored which genes drive the changes of gene expression patterns in response to time and food intake. We applied the Granger causality test and the dynamic Bayesian network to gene expression data generated from blood samples collected at multiple time points during the course of a day. The simulation result shows that combining many short time series together is as powerful to infer Granger causality as using a single long time series. Using the Granger causality test, we identified genes that were supported as the most likely causal candidates for the coordinated temporal changes in the network. These results show that PER1 is a key regulator of the blood transcriptional network, in which multiple biological processes are under circadian rhythm regulation. The fasted and fed dynamic Bayesian networks showed that over 72% of dynamic connections are self links. Finally, we show that different processes such as inflammation and lipid metabolism, which are disconnected in the static network, become dynamically linked in response to food intake, which would suggest that increasing nutritional load leads to coordinate regulation of these biological processes. In conclusion, our results suggest that food intake has a profound impact on the dynamic co-regulation of multiple biological processes, such as metabolism, immune response, apoptosis and circadian rhythm. The results could have broader implications for the design of studies of disease association and drug response in clinical trials.

Co-expression networks are routinely used to study human diseases like obesity and diabetes. Systematic comparison of these networks between species has the potential to elucidate common mechanisms that are conserved between human and rodent species, as well as those that are species-specific characterizing evolutionary plasticity. We developed a semi-parametric meta-analysis approach for combining gene-gene co-expression relationships across expression profile datasets from multiple species. The simulation results showed that the semi-parametric method is robust against noise. When applied to human, mouse, and rat liver co-expression networks, our method out-performed existing methods in identifying gene pairs with coherent biological functions. We identified a network conserved across species that highlighted cell-cell signaling, cell-adhesion and sterol biosynthesis as main biological processes represented in genome-wide association study candidate gene sets for blood lipid levels. We further developed a heterogeneity statistic to test for network differences among multiple datasets, and demonstrated that genes with species-specific interactions tend to be under positive selection throughout evolution. Finally, we identified a human-specific sub-network regulated by RXRG, which has been validated to play a different role in hyperlipidemia and Type 2 diabetes between human and mouse. Taken together, our approach represents a novel step forward in integrating gene co-expression networks from multiple large scale datasets to leverage not only common information but also differences that are dataset-specific.

BACKGROUND: An important goal of comparative genomics is the identification of functional elements through conservation analysis. Phylo-HMM was recently introduced to detect conserved elements based on multiple genome alignments, but the method has not been rigorously evaluated. RESULTS: We report here a simulation study to investigate the power of phylo-HMM. We show that the power of the phylo-HMM approach depends on many factors, the most important being the number of species-specific genomes used and evolutionary distances between pairs of species. This finding is consistent with results reported by other groups for simpler comparative genomics models. In addition, the conservation ratio of conserved elements and the expected length of the conserved elements are also major factors. In contrast, the influence of the topology and the nucleotide substitution model are relatively minor factors. CONCLUSION: Our results provide for general guidelines on how to select the number of genomes and their evolutionary distance in comparative genomics studies, as well as the level of power we can expect under different parameter settings.

To dissect common human diseases such as obesity and diabetes, a systematic approach is needed to study how genes interact with one another, and with genetic and environmental factors, to determine clinical end points or disease phenotypes. Bayesian networks provide a convenient framework for extracting relationships from noisy data and are frequently applied to large-scale data to derive causal relationships among variables of interest. Given the complexity of molecular networks underlying common human disease traits, and the fact that biological networks can change depending on environmental conditions and genetic factors, large datasets, generally involving multiple perturbations (experiments), are required to reconstruct and reliably extract information from these networks. With limited resources, the balance of coverage of multiple perturbations and multiple subjects in a single perturbation needs to be considered in the experimental design. Increasing the number of experiments, or the number of subjects in an experiment, is an expensive and time-consuming way to improve network reconstruction. Integrating multiple types of data from existing subjects might be more efficient. For example, it has recently been demonstrated that combining genotypic and gene expression data in a segregating population leads to improved network reconstruction, which in turn may lead to better predictions of the effects of experimental perturbations on any given gene. Here we simulate data based on networks reconstructed from biological data collected in a segregating mouse population and quantify the improvement in network reconstruction achieved using genotypic and gene expression data, compared with reconstruction using gene expression data alone. We demonstrate that networks reconstructed using the combined genotypic and gene expression data achieve a level of reconstruction accuracy that exceeds networks reconstructed from expression data alone, and that fewer subjects may be required to achieve this superior reconstruction accuracy. We conclude that this integrative genomics approach to reconstructing networks not only leads to more predictive network models, but also may save time and money by decreasing the amount of data that must be generated under any given condition of interest to construct predictive network models.

MOTIVATION: There is a very large and growing level of effort toward improving the platforms, experiment designs, and data analysis methods for microarray expression profiling. Along with a growing richness in the approaches there is a growing confusion among most scientists as to how to make objective comparisons and choices between them for different applications. There is a need for a standard framework for the microarray community to compare and improve analytical and statistical methods. RESULTS: We report on a microarray data set comprising 204 in-situ synthesized oligonucleotide arrays, each hybridized with two-color cDNA samples derived from 20 different human tissues and cell lines. Design of the approximately 24 000 60mer oligonucleotides that report approximately 2500 known genes on the arrays, and design of the hybridization experiments, were carried out in a way that supports the performance assessment of alternative data processing approaches and of alternative experiment and array designs. We also propose standard figures of merit for success in detecting individual differential expression changes or expression levels, and for detecting similarities and differences in expression patterns across genes and experiments. We expect this data set and the proposed figures of merit will provide a standard framework for much of the microarray community to compare and improve many analytical and statistical methods relevant to microarray data analysis, including image processing, normalization, error modeling, combining of multiple reporters per gene, use of replicate experiments, and sample referencing schemes in measurements based on expression change. AVAILABILITY/SUPPLEMENTARY INFORMATION: Expression data and supplementary information are available at http://www.rii.com/publications/2003/HE_SDS.htm