A global pathway crosstalk network.

MOTIVATION: Given the complex nature of biological systems, pathways often need to function in a coordinated fashion in order to produce appropriate physiological responses to both internal and external stimuli. Therefore, understanding the interaction and crosstalk between pathways is important for understanding the function of both cells and more complex systems. RESULTS: We have developed a computational approach to detect crosstalk among pathways based on protein interactions between the pathway components. We built a global mammalian pathway crosstalk network that includes 580 pathways (covering 4753 genes) with 1815 edges between pathways. This crosstalk network follows a power-law distribution: P(k) approximately k(-)(gamma), gamma = 1.45, where P(k) is the number of pathways with k neighbors, thus pathway interactions may exhibit the same scale-free phenomenon that has been documented for protein interaction networks. We further used this network to understand colorectal cancer progression to metastasis based on transcriptomic data. CONTACT: yong.2.li@gsk.com SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

The role of positive selection in determining the molecular cause of species differences in disease.

BACKGROUND: Related species, such as humans and chimpanzees, often experience the same disease with varying degrees of pathology, as seen in the cases of Alzheimer's disease, or differing symptomatology as in AIDS. Furthermore, certain diseases such as schizophrenia, epithelial cancers and autoimmune disorders are far more frequent in humans than in other species for reasons not associated with lifestyle. Genes that have undergone positive selection during species evolution are indicative of functional adaptations that drive species differences. Thus we investigate whether biomedical disease differences between species can be attributed to positively selected genes. RESULTS: We identified genes that putatively underwent positive selection during the evolution of humans and four mammals which are often used to model human diseases (mouse, rat, chimpanzee and dog). We show that genes predicted to have been subject to positive selection pressure during human evolution are implicated in diseases such as epithelial cancers, schizophrenia, autoimmune diseases and Alzheimer's disease, all of which differ in prevalence and symptomatology between humans and their mammalian relatives. In agreement with previous studies, the chimpanzee lineage was found to have more genes under positive selection than any of the other lineages. In addition, we found new evidence to support the hypothesis that genes that have undergone positive selection tend to interact with each other. This is the first such evidence to be detected widely among mammalian genes and may be important in identifying molecular pathways causative of species differences. CONCLUSION: Our dataset of genes predicted to have been subject to positive selection in five species serves as an informative resource that can be consulted prior to selecting appropriate animal models during drug target validation. We conclude that studying the evolution of functional and biomedical disease differences between species is an important way to gain insight into their molecular causes and may provide a method to predict when animal models do not mirror human biology.

A comparison of statistical tests for detecting differential expression using Affymetrix oligonucleotide microarrays.

Signal quantification and detection of differential expression are critical steps in the analysis of Affymetrix microarray data. Many methods have been proposed in the literature for each of these steps. The goal of this paper is to evaluate several signal quantification methods (GCRMA, RSVD, VSN, MAS5, and Resolver) and statistical methods for differential expression (t test, Cyber-T, SAM, LPE, RankProducts, Resolver RatioBuild). Our particular focus is on the ability to detect differential expression via statistical tests. We have used two different datasets for our evaluation. First, we have used the HG-U133 Latin Square spike in dataset developed by Affymetrix. Second, we have used data from an in-house rat liver transcriptomics study following 30 different drug treatments generated using the Affymetrix RAE230A chip. Our overall recommendation based on this study is to use GCRMA for signal quantification. For detection of differential expression, GCRMA coupled with Cyber-T or SAM is the best approach, as measured by area under the receiver operating characteristic (ROC) curve. The integrated pipeline in Resolver RatioBuild combining signal quantification and detection of differential expression is an equally good alternative for detecting differentially expressed genes. For most of the differential expression algorithms we considered, the performance using MAS5 signal quantification was inferior to that of the other methods we evaluated.

Pazopanib efficacy in renal cell carcinoma: evidence for predictive genetic markers in angiogenesis-related and exposure-related genes.

PURPOSE: Pazopanib, an oral angiogenesis inhibitor, is approved for the treatment of advanced renal cell carcinoma (RCC). Response to pazopanib monotherapy varies between patients, and no validated biomarkers predictive of treatment outcome have been identified. We tested the hypothesis that this variability is partially dependent on germline genetic variants that may affect pazopanib exposure or angiogenesis pathways. PATIENTS AND METHODS: Twenty-seven functional polymorphisms within 13 genes were evaluated in 397 patients with RCC. Genetic association with progression-free survival (PFS) and objective response rate (RR) was analyzed using the Cox proportional hazards model and proportional odds model, respectively. RESULTS: Three polymorphisms in IL8 and HIF1A and five polymorphisms in HIF1A, NR1I2, and VEGFA showed nominally significant association (P ≤ .05) with PFS and RR, respectively. Compared with the wild-type AA genotype (median PFS, 48 weeks), the IL8 2767TT variant genotype showed inferior PFS (27 weeks, P = .009). The HIF1A 1790AG genotype was associated with inferior PFS and reduced RR, compared with the wild-type GG genotype (median PFS, 20 v 44 weeks; P = .03; RR, 30% v 43%, P = .02). Reductions in RR were detected for the NR1I2 -25385TT genotype, compared with the wild-type CC genotype (37% v 50%, P = .03), and for the VEGFA -1498CC genotype compared with the TT genotypes (33% v 51%). CONCLUSION: Germline variants in angiogenesis- and exposure-related genes may predict treatment response to pazopanib monotherapy in patients with RCC. If validated, these markers may explain why certain patients fail antiangiogenesis therapy and they may support the use of alternative strategies to circumvent this issue.

Plasma cytokine and angiogenic factor profiling identifies markers associated with tumor shrinkage in early-stage non-small cell lung cancer patients treated with pazopanib.

There is an unmet need for pharmacodynamic and predictive biomarkers for antiangiogenic agents. Recent studies have shown that soluble vascular endothelial growth factor receptor 2 (sVEGFR2), VEGF, and several other soluble factors may be modulated by VEGF pathway inhibitors. We conducted a broad profiling of cytokine and angiogenic factors (CAF) to investigate the relationship between baseline CAF levels, CAF changes during treatment, and tumor shrinkage in early-stage non-small cell lung cancer (NSCLC) patients treated with pazopanib, an oral angiogenesis inhibitor targeting VEGFR, platelet-derived growth factor receptor, and c-kit. Plasma samples were collected before treatment and on the last day of therapy from 33 patients with early-stage NSCLC participating in a single-arm phase II trial. Levels of 31 CAFs were measured by suspension bead multiplex assays or ELISA and correlated with change in tumor volume. Pazopanib therapy was associated with significant changes of eight CAFs; sVEGFR2 showed the largest decrease, whereas placental growth factor underwent the largest increase. Increases were also observed in stromal cell-derived factor-1alpha, IP-10, cutaneous T-cell-attracting chemokine, monokine induced by IFN-gamma, tumor necrosis factor-related apoptosis-inducing ligand, and IFN-alpha. Posttreatment changes in plasma sVEGFR2 and interleukin (IL)-4 significantly correlated with tumor shrinkage. Baseline levels of 11 CAFs significantly correlated with tumor shrinkage, with IL-12 showing the strongest association. Using multivariate classification, a baseline CAF signature consisting of hepatocyte growth factor and IL-12 was associated with tumor response to pazopanib and identified responding patients with 81% accuracy. These data suggest that CAF profiling may be useful for identifying patients likely to benefit from pazopanib, and merit further investigation in clinical trials.

Inferring pathways from gene lists using a literature-derived network of biological relationships.

MOTIVATION: A number of omic technologies such as transcriptional profiling, proteomics, literature searches, genetic association, etc. help in the identification of sets of important genes. A subset of these genes may act in a coordinated manner, possibly because they are part of the same biological pathway. Interpreting such gene lists and relating them to pathways is a challenging task. Databases of biological relationships between thousands of mammalian genes can help in deciphering omics data. The relationships between genes can be assembled into a biological network with each protein as a node and each relationship as an edge between two proteins (or nodes). This network may then be searched for subnetworks consisting largely of interesting genes from the omics experiment. The subset of genes in the subnetwork along with the web of relationships between them helps to decipher the underlying pathways. Finding such subnetworks that maximally include all proteins from the query set but few others is the focus for this paper. RESULTS: We present a heuristic algorithm and a scoring function that work well both on simulated data and on data from known pathways. The scoring function is an extension of a previous study for a single biological experiment. We use a simple set of heuristics that provide a more efficient solution than the simulated annealing method. We find that our method works on reasonably complex curated networks containing approximately 9000 biological entities (genes and metabolites), and approximately 30,000 biological relationships. We also show that our method can pick up a pathway signal from a query list including a moderate number of genes unrelated to the pathway. In addition, we quantify the sensitivity and specificity of the technique.

A comparison of statistical methods for analysis of high density oligonucleotide array data.

MOTIVATION: Gene expression profiling has become an invaluable tool in functional genomics. A wide variety of statistical methods have been employed to analyze the data generated in experiments using Affymetrix GeneChip microarrays. It is important to understand the relative performance of these methods in terms of accuracy in detecting and quantifying relative gene expression levels and changes in gene expression. RESULTS: Three different analysis approaches have been compared in this work: non-parametric statistical methods implemented in Affymetrix Microarray Analysis Suite v5.0 (MAS5); an error-modeling based approach implemented in Rosetta Resolver v3.1; and an intensity-modeling approach implemented in dChip v1.1. A Latin Square data set generated and made available by Affymetrix was used in the comparison. All three methods-Resolver, MAS5 and the version of dChip based on the difference between perfect match and mismatch intensities-perform well in quantifying gene expression. Presence calls made by MAS5 and Resolver perform well at high concentrations, but they cannot be relied upon at low concentrations. The performance of Resolver and MAS5 in detecting 2-fold changes in transcript concentration is superior to that of dChip. At a comparable false positive rate, Resolver and MAS5 are able to detect many more true changes in transcript concentration. Estimated fold changes calculated by all the methods are biased below the true values.