Pathway Analysis in Drug Development
The drive of the drug discovery industry toward personalized medicine necessitates the development of a pathway analysis infrastructure.
Azat Badretdinov, Anton Yuryev
Ariadne Genomics Inc, Rockville, MD 20874 USA
 |
The term “pathway analysis” was often used to describe the studies of dynamics, stability and sensitivity of such pathways to external signals. In the following sections, we refer to these studies as “pathway simulations,” since pathway analysis is increasingly used to interpret high-throughput experiments that measure the in-vivo abundance of various biological molecules. Methods include measuring mRNA abundance using gene expression microarrays (Elvidge, 2006; Stoughton 2005), metabolomics experiments measuring endogenous and drug metabolite concentration (Weckwerth, 2005), proteomics experiments measuring protein levels (Stubberfield, 1999), studies of protein phosphorylation and protein-protein interactions by protein arrays, mass spectrometry or yeast two-hybrid screen (Flory, 2003, Rual, 2005). All of these approaches generate thousands of data points in a single run, making manual analysis impossible.
One goal of pathway analysis is to find an explanation for an observed molecular profile by predicting activated pathways that cause it. Most approaches use known molecular interactions to calculate pathways, but some try to infer novel interactions directly from the profiling data. Information about activated pathways can be used to select known drugs for personalized therapy, to select and prioritize potential drug targets to develop new drugs and to evaluate the efficacy of a drug candidate or to predict drug side effects and toxicity. This article reviews existing applications and explores additional possibilities for pathway analysis in drug discovery. We argue that the drive of the drug discovery industry toward personalized medicine necessitates the development of a pathway analysis infrastructure.
Since pathway analysis cannot be performed without the appropriate software tools, we start with reviewing available software technologies.
Databases for Pathway Analysis
Two major sources of molecular interaction data exist: peer-reviewed scientific articles and high-throughput experiments mentioned previously. Several databases of molecular interaction have been developed using manual curation. Public manually curated databases include: BIND, KEGG, DIP, HPRD, and Reactome. Commercial manually-curated databases have been developed by Ingenuity Systems, Jubilant Biosystems, Molecular Connections and GeneGo. Because these databases provide highly accurate data, they suffer from slow and expensive data accumulation.
The exponential explosion of the number of articles describing molecular interactions has led to the development of automatic extraction of molecular interactions from scientific publications. These methods extract information with some error rate but also provide an inexpensive and fast alternative to manual curation. Ariadne Genomics has created a text-to-knowledge technology called Medscan, which extracts more than 1,000,000 relations from 14,000,000 PubMed abstracts and 43 full-text journals with about a 10% error rate (Daraselia, 2004). Other text-mining technologies are available from Genomatix Inc.(Bibliosphere), TEMIS (Insight Discoverer), Nervana (Semantic Medline), European Bioinformatics Institute and Agilent.
Current knowledge about molecular interaction is incomplete. For example, the ResNet 4.0 database from Ariadne Genomics combines data automatically extracted from the literature, curated interactions from the public domain and results of high-throughput experiments. It contains relations for 15,809 proteins, which is less than 50% of the Human Genome. The three previously described methods of data accumulation—manual curation, automatic text-mining and high-throughput experiments—will most likely be used in the foreseeable future.
 |
Pathway analysis allows decomposition of experimental data into a set of known pathways and biological processes. Unfortunately, this approach was hampered by the scarcity of pathway data. The database of metabolic pathways for human tissues was constructed only last year (Romero, 2005). Currently, known signaling pathways are scattered among various public and commercial manually curated databases: KEGG has approximately 24 regulatory pathways; STKE database has about 80 signaling pathways; Biocarta contains about 250 pathways for human, mouse and rat organisms; PathArt database (Jubilant) contains 696 disease-associated pathways describing various biological processes in 26 diseases. Cambridge Cell Networks is undertaking the effort to generate toxicity-associated pathways.
The following calculation shows that building a comprehensive pathway database is impossible using manual curation. More than 520 signaling ligands in the Human Genome and 200 human organism tissues exist. About 100,000 signaling pathways for different tissues are estimated. Ariadne Genomics’ protein-disease database contains 10,694 proteins associated in the literature with one of 1,779 different diseases. Combining this number with the average number of processes for one disease from the PathArt database predicts more than 46,000 disease-associated pathways. These numbers will undoubtedly grow, ultimately making manual pathway building an arduous effort. Thus, building a pathway database obstructs efficient pathway analysis. This bottleneck can only be resolved by the development of computer algorithms for pathway reconstruction. Such algorithms are being developed using a combination of known regulatory networks, physical interactions, and gene expression data (Figure 1).
Software tools for navigating the database of molecular interactions to analyze molecular profiling data.
The identification of molecular signatures for diseases and toxic states became possible due to the recent advances in gene expression microarray and other high-throughput technologies (Bild, 2006; Solit, 2006; Heller, 1997; Haferlach, 2003; Nicholson, 2002). These signatures are identified by a cohort of algorithms that find molecules showing statistically significant variations between different samples (Hariharan, 2003; Draghici, 2002) and groups of molecules having similar changes across multiple samples using statistical clustering (Shannon, 2003). These algorithms find statistically significant sets of genes, yet they do not provide insights into why the changes have taken place.
Network database navigation tools can find a minimal set of regulators responsible for most of changes in molecular profiles and further reveal why the regulator activity could be modified. For example, a typical microarray experiment finds more than 1,000 differentially expressed genes, but their expression is driven in theory by a far smaller number of transcription factors. Most commercial companies previously mentioned provide such tools, with varying degrees of sophistication. An open source Cytoscape package is also available. Standard algorithms for network navigation (Moore, 1959; Nikitin, 2003) enable network expansion by finding the shortest path between the database entities, common regulators and targets. Pathway analysis using these algorithms is still laborious and time consuming, yet it is impossible without them.
The representation of the molecular profile as a set of major regulators reduces the complexity of the observed pattern and simplifies further analysis. Complexity can be reduced further by finding molecular complexes in the interaction network among differentially expressed genes (Ideker, 2002). The set of major regulators, in combination with their downstream targets exhibiting differential profiles, is interpreted as a list of affected biological processes (Calvano, 2005). Note that the reduction of complexity is also a fundamental task for developing efficient layout algorithms by simplifying large pathway diagram visualization (Ju, 2003; Kohn, 2001).
Another major application for current pathway analysis tools is biomarker discovery and optimization. For instance, the most convenient biomarkers are secreted proteins and metabolites. The statistically significant expression level variations are usually identified using microarray technology that cannot detect changes in secretion or signaling through protein modification and chemical reactions. Nevertheless, the optimal biomarkers can be identified as downstream targets of the differentially expressed genes.
Pathway reconstruction for normal, disease and toxic state for drug target selection.
Molecular signatures are powerful diagnostic tools, but the selection of a drug target requires complete knowledge of the regulatory interaction in a pathway. It is necessary to understand the downstream/upstream hierarchy that causes a molecular signature. The expression signatures for known pathways have been already proposed to improve cancer treatment (Solit, 2006). However, it is impossible to accumulate the collection of such signatures for all signaling pathways in all human tissues. The dependency of molecular profiles on a microarray platform (Jarvinen, 2004) further necessitates the development of methods for platform-independent pathway analysis.
Pathway reconstruction from molecular signatures is a crucial computational problem yet to be addressed. In principle, every observed molecular pattern can be explained by a minimal set of upstream regulators. The path from the regulator(s) to the proteins comprising a molecular signature is called an “activated pathway.” The activity of these regulators can be altered due to mutation, aberrant signaling from a tissue damage or environmental factors. Perhaps the biggest challenge for pathway calculation is the cross-talk between pathways (Velloso, 2006; Linask, 2005; Dailey, 2005; Vilarem, 2004). The interactions occur due to the existence of common protein members between two pathways (Izumi, 2006) and are presented schematically in Figure 2. The individual genotype information may help to overcome some of this complexity and has to be taken into consideration by such algorithms. Once major regulators are identified, the drug target selection is reduced to finding druggable proteins among or upstream of the regulators (Herrmann, 2001; Hopkins and Groom, 2002; Davies, 2005).
A strong algorithmic base for metabolic pathway reconstruction exists (Kell 2004); however, metabolic enzymes only recently have been suggested as drug targets (Boros 2004). The fundamental role these enzymes play in every tissue may hamper their use in drug development, due to potential side-effects. Nevertheless, metabolic profiling and reconstruction in tumor cells can clearly help in diagnosing and understanding the disease mechanism.
Pathway Analysis for drug target and lead compound validation and for predictive toxicology.
 |
Pathway analysis allows the prediction drug side effects. First, the drug target itself can participate in another pathway or in the same pathway but in a different tissue. The interrogation of a pathway database to find pathways containing a drug target linked to the analysis of affected biological processes can provide invaluable insight on potential side effects. Second, methods of in silico rational structure-based drug design can predict other proteins potentially binding to lead (Gane, 2000; Kirkpatrick, 1999). These proteins must be evaluated for potential side effects in the same manner as the drug target itself.
The example of combining computational drug design with pathway analysis exists for the P450 system (Ekins, 2005). The method uses a quantitative structure activity relationship to predict compound binding to cytochrome P450 proteins. Empirical molecular descriptors are used to predict drug metabolites produced by P450 enzymes (Balakin, 2004). The effects of metabolites and P450 activation are then inferred on other signaling and metabolic pathways using molecular interaction data. The prediction of side effects by this approach is possible due to the large collection of xenobiotic reactions known for the P450 system. Despite the limited use, it demonstrates the validity of the in silico prediction of the drug action.
Pathway simulation and drug design.
One of the most straightforward applications of pathway analysis for drug discovery is predicting the effect of a drug on pathway behavior by simulating the kinetics of concentrations of the pathway components. This can be done at the highest level of detail by numerically solving ordinary differential equations (ODE) (Kofahl, 2004) or stochastic modeling (McCollum, 2006). Simplified approaches include a steady concentration assumption in the method of minimization of metabolic adjustment (MOMA) flux balance analysis (Segre, 2002), treating each component of the kinetic models as being in one or several discrete states. Examples of such systems are Petri Nets (Kuffner, 2000), Boolean networks (Akutsu, 2000), S-systems (Akutsu, 2000), and Bayesian networks (Yoshida, 2005). Currently, one of the significant disadvantages of the ODE-based method is the scarcity of kinetic parameters in protein biochemistry. On the other hand, surprisingly, some kinetic models are reported to be quite insensitive to several degrees of magnitude of variation of kinetic parameters (von Dassow, 2000). This evidence opens a possibility for a design of a set of universal kinetic models that can be suitable for drug action analysis. Kinetic parameters, in this case, can be estimated using information about kinetic constants for orthologous and paralogous proteins or directly from protein structure. Certainly, attempts to develop simulation models relevant to drug discovery continue (Christopher 2004; d'Alché-Buc 2004).
Pathway Analysis for design of promiscuous drugs and selective drug mixtures.
Pharmaceutical research has produced many highly selective chemical protein inhibitors. Yet, the number of drugs approved by the Food and Drug Administration (FDA) last year has decreased. The FDA approves now only one out of 5,000 drugs (Davies, 2006). The failures of blockbuster drugs such as Vioxx, Bextra and Fen-Phen has prompted a re-examination of the focus of the pharmaceutical industry on increasing drug selectivity (Frantz, 2005; Mencher, 2005; Arteaga, 2003). Support for the development of promiscuous drugs and drug mixtures comes from homeopathic medicine, which is based entirely on the mixtures of herbal extracts, and from the promiscuity of widely known drugs such as aspirin (Mencher, 2005). Current knowledge of modern molecular biology postulates that most human diseases are complex; i.e., their genetic predisposition is caused by mutations in multiple genes. This fact alone implies the involvement of multiple pathways in one disease. In addition, regulatory pathways have many redundant signaling paths, and they tightly intertwine in vivo (Berenbaum, 2004; Gong, 2005; Ferreira, 2005). Thus, finding a cure by completely blocking an aberrant signaling in a one pathway probably will affect multiple targets. Several targets in one pathway, multiplied by a number of pathways involved in a disease, provide an estimate for the number of selective inhibitors necessary to provide an effective treatment.
In principle, drug promiscuity can be determined by using a combination of structure-based rational drug design methods: molecular docking (Mohan, 2005) and protein structure homology modeling (Muegge, 2004). The calculation of the pleothropic effect of a promiscuous drug is a task for pathway analysis. These effects can be calculated by propagating a signal from multiple drug targets and calculating the cumulative effect. The accuracy of this prediction depends entirely on the completeness and accuracy of the regulatory network available in the database. Experimental approaches for combinatory medicine are already being developed (Borisy, 2003). Experimentally-determined effective drug combinations should improve our understanding about pathway interaction in vivo.
The most obvious approach for the rational design of drug mixtures is finding compounds that compensate for toxic side effects of a primary drug. Indeed, due to the druggability constraints, it may not be possible to find a drug target without side effects. In this case, the side effects have to be down-regulated by other compounds. Examples of such drug cocktails or combinatory therapy already exist for cancer and HIV (Mahesh, 2005; Zhou, 2004; Doerfler, 2002). In principle, a prediction of either adverse or synergistic positive effects of interactions between known drugs can also be done by signal propagation along the regulatory network.
Towards comprehensive prediction of drug action
The pathway analysis field is experiencing booming growth. It can be viewed as the last frontier in the transformation of molecular biology, from the art of cloning a single gene to a science driven by numbers and statistical computations. Once developed, pathway analysis should accomplish the ultimate goal of rational drug design: to quickly develop safe drugs with reliably predicted effects.
Current experimental approaches to drug efficacy and toxicity assessment are performed via expensive and often unreliable experiments using animal models and clinical trials. True in silico rational drug design must include reliable predictions, which can only be made using pathway analysis infrastructure and algorithms.
One of the remaining central questions is whether a cell transmits the information through fairly isolated pathways, or whether the whole signaling network must be considered. The answer to this question is related to the difference between “system biology” and “pathway analysis” methods. Indeed, many issues raised here are also mentioned by other authors in the context of the “system biology” approach (Ekins, 2006, Butcher, 2005). If history is any guide, the answer to this question, as in many other questions in biology, is somewhere in the middle. For example, it may be possible to represent signaling in a healthy cell as a collection of rather independent modules. These modules may become increasingly “disheveled” in a disease state. Alternatively, if the propagation of a signal can be calculated only by considering the entire regulatory network of the healthy cell, a disease may activate distinct sub-networks that can be considered as pathways. Clearly, current state-of-the-art pathway analysis software allows the integration of individual pathways into one global network and the representation of the pathways as sub-networks. Whether these sub-networks function independently and what portion of the global network is usually activated under different conditions remain open questions that can be answered only using pathway analysis tools.
The fact that effective cures for most modern diseases will require the personalized medicine approach is gaining acceptance. High-throughput genotyping methods provide initial understanding of mutations contributing to the disease state. Molecular profiling enables quick assessment of the current condition of the patient. Using this information, pathway analysis provides a means to swiftly calculate possible drug targets and formulate an effective cure. Phrases such as “rational drug design” and “personalized medicine” are already included in the lexicon of serious scientific publications. The emerging workflow for personalized medicine consists of the following stages:
1. Disease diagnosis using molecular profiling and biomarkers;
2. Calculating the activated pathways in the disease state and molecular causes for disease using genotype information and database of molecular interactions;
3. Selection of known drugs or de-novo design of new drugs based on the information about pathways activated in disease and in normal state, and
4. Formulation of the personalized drug recipes maximized for efficacy and minimized for toxicity. Pathway analysis has a prominent role in every stage.
Acknowledgements.
We thank Drs. Ilya Mazo, Yaraslav Ispolatov and Andrey Sivachenko for valuable comments about the manuscript. We thank Lori Wilson for the language review.
References
Aboul-Fadl T. (2005) Antisense oligonucleotides: the state of the art. Curr Med Chem.;12(19):2193-214.
Alban A, Chanan-Khan, Padmanabhan S, Stein L, Panzarella J, Kena C. Miller, and Hawthorne L. (2005) Validating Molecular Targets of Thalidomide in CLL: Net Effect of Increased Apoptosis through the Intrinsic Pathway and down Regulation of NF-kB Signaling- Validation Using Gene Expression Profile from the Phase I/II Clinical Trial of Thalidomide and Fludarabine. Blood (ASH Annual Meeting Abstracts); 106: 5043
Akutsu T, Miyano S, Kuhara S. (2000) Inferring qualitative relations in genetic networks and metabolic pathways. Bioinformatics; 16:727-34.
Arteaga CL. (2003) Is One Promiscuous Drug against Multiple Targets Better than Combinations of Molecule-specific Drugs? Clinical Cancer Research Vol. 9, 1231-1232
Bader, J.S., Chaudhuri, A., Rothberg, J.M. and Chant, J. (2004) Gaining confidence in high-throughput protein interaction networks. Nat. Biotechnol. 22, 78–85.
Balakin KV, Ekins S, Bugrim A, Ivanenkov YA, Korolev D, Nikolsky YV, Ivashchenko AA, Savchuk NP, Nikolskaya T. (2004) Quantitative structure-metabolism relationship modeling of metabolic N-dealkylation reaction rates. Drug Metab Dispos. 32(10):1111-20.
Berenbaum F. (2004) Signaling transduction: target in osteoarthritis. Curr Opin Rheumatol. Sep;16(5):616-22.
Bild AH, Yao G, Chang JT, Wang Q, Potti A, Chasse D, Joshi MB, Harpole D, Lancaster JM, Berchuck A, Olson JA Jr, Marks JR, Dressman HK, West M, Nevins JR. (2006) Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature 19;439(7074):353-7
Borisy AA., et al. (2003) Systematic discovery of multicomponent therapeutics. Proc. Natl. Acad. Sci. USA 100, 7977–7982.
Boros LJ, Serkova NJ, Cascante MS, Lee WNP. (2004) Use of metabolic pathway flux information in targeted cancer drug design. Drug Discovery Today: Therapeutic Strategies Vol. 1, No. 4
Butcher EC. (2005) Can cell systems biology rescue drug discovery? Nat Rev Drug Discov. 4(6):461-7.
Calvano SE, Xiao W, Richards DR, Felciano RM, Baker HV, Cho RJ, Chen RO, Brownstein BH, Cobb JP, Tschoeke SK, Miller-Graziano C, Moldawer LL, Mindrinos MN, Davis RW, Tompkins RG, Lowry SF. (2005) Inflamm and Host Response to Injury Large Scale Collab. Res. Program. A network-based analysis of systemic inflammation in humans.Nature. 13;437(7061):1032-7.
Catley MC, Chivers JE, Holden NS, Barnes PJ, Newton R. (2005) Validation of IKK beta as therapeutic target in airway inflammatory disease by adenoviral-mediated delivery of dominant-negative IKK beta to pulmonary epithelial cells. Br J Pharmacol. May;145(1):114-22.
Cejka D, Losert D, Wacheck V. (2006) Short interfering RNA (siRNA): tool or therapeutic? Clin Sci (Lond) 110(1):47-58.
Christopher R, Dhiman A, Fox J, Gendelman R, Haberitcher T, Kagle D, Spizz G, Khalil IG, Hill C. (2004) Data-driven computer simulation of human cancer cell. Ann N Y Acad Sci. 1020:132-53.
Dailey L, Ambrosetti D, Mansukhani A, Basilico C. (2005) Mechanisms underlying differential responses to FGF signaling. Cytokine Growth Factor Rev. 16(2):233-47.
d'Alché-Buc F, Lahaye PJ, Perrin BE, Ralaivola L, Vujasinovic T, Mazurie A, Bottani S. (2004) A dynamic model of gene regulatory networks based on inertia principle. In U. Seiffert editors, Bioinformatics using Computational Intelligence Paradigms, Springer
Daraselia N, Yuryev A, Egorov S, Novichkova S, Nikitin A, Mazo I. (2004) Extracting human protein interactions from MEDLINE using a full-sentence parser. Bioinformatics. 22;20(5):604-11.
von Dassow G, Meir E, Munro EM, Odell GM. (2000) The segment polarity network is a robust developmental module. Nature. 406:188-92.
Davies K. “Cracking the 'Druggable Genome” (2005) BioIT World, (http://www.bio-itworld.com/archive/100902/firstbase.html)
Davies K, Counting the Cost of Drug Discovery, (2006) BIO-IT World, http://www.bio-itworld.com/archive/071102/firstbase.html
Deane, C.M., Salwinski, L., Xenarios, I. and Eisenberg, D. (2002) Protein interactions: two methods for assessment of the reliability of high throughput observations. Mol. Cell Proteomics 1, 349–356.
Doerfler RE. (2002) Sweetening the toxic cocktail. Managing the side effects of HIV medications. Adv Nurse Pract. 10(5):52-4, 57-8, 103.
Draghici S. (2002) Statistical intelligence: effective analysis of high-density microarray data.Drug Discov Today. 1;7(11 Suppl):S55-63.
Elvidge G. (2006) Microarray expression technology: from start to finish. Pharmacogenomics. 7(1):123-34.
Ekins S. (2006) Systems-ADME/Tox: resources and network approaches. J Pharmacol Toxicol Methods. 53(1):38-66.
Ekins S, Andreyev S, Rybadov A, Kirillov E, Rakhmatulin EA, Sorokina S, Bugrim A, Nikolskaya T. (2005) A combined approach to drug metabolism and toxicity assessment. Drug Metab Dispos.
Ferreira FJ, Kieber JJ. (2005) Cytokinin signaling. Curr Opin Plant Biol. 8(5):518-25
Flory MR, Aebersold R. (2003) Proteomic approaches for the identification of cell cycle-related drug targets.Prog Cell Cycle Res.5:167-71.
Frantz S. (2005) Drug discovery: Playing dirty Nature 437, 942-943
Gane PJ, Dean PM. (2000) Recent advances in structure-based rational drug designs. Curr Opin Struct Biol. 10(4):401-4.
Gielen SCJP, Kuhne LCM, Ewing PC, Blok LJ, and Burger CW (2005) Tamoxifen treatment for breast cancer enforces a distinct gene-expression profile on the human endometrium: an exploratory study. Endocr. Relat. Cancer, 12: 1037 - 1049.
Giot, L., et al. (2003) A protein interaction map of Drosophila melanogaster. Science 302, 1727–1736.
Goldberg, D.S. and Roth, F.P. (2003) Assessing experimentally derived interactions in a small world. Proc. Natl. Acad. Sci. USA 100, 4372–4376.
Gong Y, Zhang Z. (2005) Alternative signaling pathways: when, where and why? FEBS Lett. 10;579(24):5265-74
Haferlach T, Kohlmann A, Kern W, Hiddemann W, Schnittger S, Schoch C. (2003) Gene expression profiling as a tool for the diagnosis of acute leukemias. Semin Hematol. 40(4):281-95.
Hariharan R. (2003) The analysis of microarray data. Pharmacogenomics. 4(4):477-97
Heller RA, Schena M, Chai A, Shalon D, Bedilion T, Gilmore J,Woolley DE, Davis RW (1997) Discovery and analysis of inflammatory disease-related genes using cDNA microarrays. Proc NatlAcad Sci USA 94: 2150–2155
Herrmann JL, Rastelli L, Burgess CE, Fernandez EE, Rothberg BE, Rothberg JM, Shimkets RA. (2001) Implications of oncogenomics for cancer research and clinical oncology. Cancer J. 7(1):40-51.
Hopkins AL, Groom CR. (2002) The druggable genome. Nat Rev Drug Discov. 1(9):727-30.
Ideker T, Ozier O, Schwikowski B, Siegel AF. (2002) Discovering regulatory and signalling circuits in molecular interaction networks. Bioinformatics18 Suppl 1:S233-40.
Izumi M, Masaki M, Hiramoto Y, Sugiyama S, Kuroda T, Terai K, Hori M, Kawase I, Hirota H. (2006) Cross-talk between bone morphogenetic protein 2 and leukemia inhibitory factor through ERK 1/2 and Smad1 in protection against doxorubicin-induced injury of cardiomyocytes. Mol Cell Cardiol.
Jarvinen AK, Hautaniemi S, Edgren H, Auvinen P, Saarela J, Kallioniemi OP, Monni O. (2004) Are data from different gene expression microarray platforms comparable? Genomics. 83(6):1164-8.
Ju BH, Han K. (2003) Complexity management in visualizing protein interaction networks.Bioinformatics19 Suppl 1:i177-9.
Kell DB. (2004) Metabolomics and systems biology: making sense of the soup. Curr Opin Microbiol. 7(3):296-307.
Kirkpatrick DL, Watson S, Ulhaq S. (1999) Structure-based drug design: combinatorial chemistry and molecular modeling. Comb Chem High Throughput Screen. 2(4):211-21.
Kofahl B, Klipp E. (2004) Modelling the dynamics of the yeast pheromone pathway. Yeast, 21:831-50.
Kohn KW. (2001) Molecular interaction maps as information organizers and simulation guides.Chaos. 11(1):84-97.
Kuffner R, Zimmer R, Lengauer T. (2000) Pathway analysis in metabolic databases via differential metabolic display (DMD). Bioinformatics 16:825-36.
Kung C, Shokat KM. (2005) Small-molecule kinase-inhibitor target assessment. Chembiochem. Mar;6(3):523-6
Linask KK, Manisastry S, Han M. (2005) Cross talk between cell-cell and cell-matrix adhesion signaling pathways during heart organogenesis: implications for cardiac birth defects.Microsc Microanal. 11(3):200-8.
Mahesh R, Perumal RV, Pandi PV. (2005) Cancer chemotherapy-induced nausea and vomiting: role of mediators, development of drugs and treatment methods. Pharmazie. 60(2):83-96.
McCollum JM, Peterson GD, Cox CD, Simpson ML, Samatova NF. (2006) The sorting direct method for stochastic simulation of biochemical systems with varying reaction execution behavior. Comput Biol Chem. 30:39-49.
Mencher SK and Wang LG. (2005) Promiscuous drugs compared to selective drugs (promiscuity can be a virtue). BMC Clinical Pharmacology, 5:3
von Mering, C., et al. (2005) STRING: known and predicted protein–protein associations, integrated and transferred across organisms. Nucleic Acids Res. 33, D433–D437.
Mohan V, Gibbs AC, Cummings MD, Jaeger EP, DesJarlais RL. (2005) Docking: successes and challenges. Curr Pharm Des.11(3):323-33
Moore EF (1959) The Shortest Path Through a Maze.Proc. International Symposium on the Theory of Switching,Part II, Vol. 30 of “The Annals of the ComputationLaboratory of Harvard University”, Cambridge, MA,Harvard University Press.
Muegge I, Enyedy IJ. (2004) Virtual screening for kinase targets. Curr Med Chem. 11(6):693-707.
Nicholson JK, Connelly J, Lindon JC, Holmes E. (2002) Metabonomics: a platform for studying drug toxicity and gene function.Nat Rev Drug Discov. 1(2):153-61.
Nikitin A., Egorov S., Daraselia N. and Mazo I. (2003) Pathway studio - the analysis and navigation of molecular networks. Alexander Bioinformatics Vol. 19 no. 0, pages 1-3.
Romero P., Wagg J, Green ML, Kaiser D, Krummenacker M, Karp PD (2005) Computational prediction of human metabolic pathways from the complete human genome. Genome Biol.6(1): R2
Rual JF et al, (2005) Towards a proteome-scale map of the human protein-protein interaction network. Nature. 20;437(7062):1173-8.
Saito, R., Suzuki, H. and Hayashizaki, Y. (2003) Construction of reliable protein–protein interaction networks with a new interaction generality measure. Bioinformatics 19, 756–763.
Segre D, Vitkup D, Church GM. (2002) Analysis of optimality in natural and perturbed metabolic networks. Proc Natl Acad Sci U S A., 99:15112-7
Shannon W, Culverhouse R, Duncan J. (2003) Analyzing microarray data using cluster analysis.Pharmacogenomics. 4(1):41-52.
Solit DB, Garraway LA, Pratilas CA, Sawai A, Getz G, Basso A, Ye Q, Lobo JM, She Y, Osman I, Golub TR, Sebolt-Leopold J, Sellers WR, Rosen N. (2006) BRAF mutation predicts sensitivity to MEK inhibition. Nature 19;439(7074):358-62.
Sprinzak E, Sattath S, Margalit H, (2003) How Reliable are Experimental Protein-Protein Interaction Data? Journal of Molecular Biology, 327(5):919-923.
Stoughton RB. (2005) Applications of DNA microarrays in biology. Annu Rev Biochem.;74:53-82.
Stubberfield CR, Page MJ. (1999) Applying proteomics to drug discovery. Expert Opin Investig Drugs. Jan;8(1):65-70.
Taylor MF. (1999) Antisense oligonucleotides for target validation and gene function determination. IDrugs. 2(8):777-81
Velloso LA, Folli F, Perego L, Saad MJ. (2006) The multi-faceted cross-talk between the insulin and angiotensin II signaling systems. Diabetes Metab Res Rev.
Weckwerth W, Morgenthal K. (2005) Metabolomics: from pattern recognition to biological interpretation. Drug Discov Today. 10(22):1551-8.
Wood JN, Boorman J. (2005) Voltage-gated sodium channel blockers; target validation and therapeutic potential. Curr Top Med Chem. 5(6):529-37
Yoshida R, Imoto S, Higuchi T. (2005) Estimating Time-Dependent Gene Networks from Time Series Microarray Data by Dynamic Linear Models with Markov Switching. Proc IEEE Comput Syst Bioinform Conf. 4:289-298.
Yuryev A, Mulyukov Z, Kotelnikova E, Maslov S, Egorov S, Nikitin A, Daraselia N, Mazo I, (2006) Automated pathway building in Biological Association networks. BMC Bioinformatics, in press
Zambrowicz BP, Turner CA, Sands AT. (2003) Predicting drug efficacy: knockouts model pipeline drugs of the pharmaceutical industry. Curr Opin Pharmacol.3(5):563-70
Zhou H, Tong Z, McLeod JF. (2004) "Cocktail" approaches and strategies in drug development: valuable tool or flawed science? J Clin Pharmacol. 44(2):120-34.
Figures
Figure 1. Using relations from ResNet 4.0 database we constructed 1,003 disease-associated pathways and 5,692 tissue-specific regulome pathways. This indicates that a sufficient amount of data exists in the literature in order to initiate the development of algorithms for automatic pathway reconstruction (Yuryev et al, 2006). The quality of the information extracted by Medscan appears sufficient for such algorithm development. Figure 1A shows the IL-10 pathway that was build automatically, and Figure 1B shows the IL-10 pathway built by manual curation. The proteins common for both pathways have blue halos.
Figure 1A
Figure 1B
Figure 2. Examples of signaling pathway cross-talk and redundancy. TGFBR1, TGFBR2, and TGFBR3 belong to TGF-beta receptors superfamily. AMHR2 is anti-Mullerian hormone receptor 2 that couples with ACVR1 (activin receptor) for AMH (gonadal tumor suppressor) signaling. The proteins participating in both AMH and TGF-beta pathways are shown with blue halos.
