Genomics and Proteomics Engineering in Medicine and Biology (IEEE Press Series on Biomedical Engineering)

Current applications and recent advances in genomics and proteomics Genomics and Proteomics Engineering in Medicine and Biology presents a well-rounded, interdisciplinary discussion of a topic that is at the cutting edge of both molecular biology and bioengineering. Compiling contributions by established experts, this book highlights up-to-date applications of biomedical informatics, as well as advancements in genomics-proteomics areas. Structures and algorithms are used to analyze genomic data and develop computational solutions for pathological understanding. Topics discussed include: Qualitative knowledge models Interpreting micro-array data Gene regulation bioinformatics Methods to analyze micro-array Cancer behavior and radiation therapy Error-control codes and the genome Complex life science multi-database queries Computational protein analysis Tumor and tumor suppressor proteins interactions

Excerpt

CHAPTER 1

Qualitative Knowledge Models in Functional Genomics and Proteomics

MOR PELEG, IRENE S. GABASHVILI, and RUSS B. ALTMAN

1.1. INTRODUCTION

Predicting pathological phenotypes based on genetic mutations remains a fundamental and unsolved issue. When a gene is mutated, the molecular functionality of the gene product may be affected and many cellular processes may go awry. Basic molecular functions occur in networks of interactions and events that produce subsequent cellular and physiological functions. Most knowledge of these interactions is represented diffusely in the published literature, Excel lists, and specialized relational databases and so it is difficult to assess our state of understanding at any moment. Thus it would be very useful to systematically store knowledge in data structures that allow the knowledge to be evaluated and examined in detail by scientists as well as computer algorithms. Our goal is to develop technology for representing qualitative, noisy, and sparse biological results in support of the eventual goal of fully accurate quantitative models.

In a recent paper, we described an ontology that we developed for modeling biological processes. Ontologies provide consistent definitions and interpretations of concepts in a domain of interest (e.g., biology) and enable software applications to share and reuse the knowledge consistently. Ontologies can be used to perform logical inference over the set of concepts to provide for generalization and explanation facilities. Our biological process ontology combines and extends two existing components: a workflow model and a biomedical ontology, both described in the methods and tools section. Our resulting framework possesses the following properties: (1) it allows qualitative modeling of structural and functional aspects of a biological system, (2) it includes biological and medical concept models to allow for querying biomedical information using biomedical abstractions, (3) it allows hierarchical models to manage the complexity of the representation, (4) it has a sound logical basis for automatic verification, and (5) it has an intuitive, graphical representation.

Our application domain is disease related to transfer ribonucleic acid (tRNA). Transfer RNA constitutes a good test bed because there exists rich literature on tRNA molecular structure as well as the diseases that result from abnormal structures in mitochondria (many of which affect neural processes). The main role of tRNA molecules is to be part of the machinery for the translation of the genetic message, encoded in messenger RNA (mRNA), into a protein. This process employs over 20 different tRNA molecules, each specific for one amino acid and for a particular triplet of nucleotides in mRNA (codon). Several steps take place before a tRNA molecule can participate in translation. After a gene coding for tRNA is transcribed, the RNA product is folded and processed to become a tRNA molecule. The tRNA molecules are covalently linked (acylated) with an amino acid to form amino-acylated tRNA (aa-tRNA). The aa-tRNA molecules can then bind with translation factors to form complexes that may participate in the translation process. There are three kinds of complexes that participate in translation: (i) an initiation complex is formed by exhibiting tRNA mimicry release factors that bind to the stop codon in the mRNA template or by a misfunctioning tRNA complexed with guanidine triphosphate (GTP) and elongation factor causing abnormal termination, and (iii) a ternary complex is formed by binding elongating aa-tRNAs (tRNAs that are acylated to amino acids other than formylmethionine) with GTP and the elongation factor EF-tu. During the translation process, tRNA molecules recognize the mRNA codons one by one, as the mRNA molecule moves through the cellular machine for protein synthesis: the ribosome. In 1964, Watson introduced the classical two-site model, which was the accepted model until 1984. In this model, the ribosome has two regions for tRNA binding, so-called aminoacyl (A) site and peptidyl (P) site. According to this model, initiation starts from the P site, but during the normal cycle of elongation, each tRNA enters the ribosome from the A site and proceeds to the P site before exiting into the cell's cytoplasm. Currently, it is hypothesized that the ribosome has at least three regions for tRNA binding: the A and P sites and an exit site (E site) through which the tRNA exits the ribosome into the cell's cytoplasm. Protein synthesis is terminated when a stop codon is reached at the ribosomal A site and recognized by a specific termination complex, probably involving factors mimicking tRNA. Premature termination (e.g., due to a mutation in tRNA) can also be observed.

When aa-tRNA molecules bind to the A site, they normally recognize and bind to matching mRNA codons—a process known as reading. The tRNA mutations can cause abnormal reading that leads to mutated protein products of translation. Types of abnormal reading include (1) misreading, where tRNA with nonmatching amino acid binds to the ribosome's A site; (2) frame shifting, where tRNA that causes frame shifting (e.g., binds to four nucleotides of the mRNA at the A site) participates in elongation; and (3) halting, where tRNA that cause premature termination (e.g., tRNA that is not acetylated with an amino acid) binds to the A site. These three types of errors, along with the inability to bind to the A site or destruction by cellular enzymes due to misfolding, can create complex changes in protein profiles of cells. This can affect all molecular partners of produced proteins in the chain of events connecting genotype to phenotype and produce a variety of phenotypes. Mutations in human tRNA molecules have been implicated in a wide range of disorders, including myopathies, encephalopathies, cardiopathies, diabetes, growth retardation, and aging. Development of models that consolidate and integrate our understanding of the molecular foundations for these diseases, based on available structural, biochemical, and physiological knowledge, is therefore urgently needed.

In a recent paper, we discussed an application of our biological process ontology to genomics and proteomics. This chapter extends the section on general computer science theories, including Petri Nets, ontologies, and information systems modeling methodologies, as well as extends the section on biological sources of information and discusses the compatibility of our outputs with popular databases and modeling environments.

The chapter is organized as follows. Section 1.2 describes the components we used to develop the framework and the knowledge sources for our model. Section 1.3 discusses our modeling approach and demonstrates our knowledge model and the way in which information can be viewed and queried using the process of translation as examples. We conclude with a discussion and conclusion.

1.2. METHODS AND TOOLS

1.2.1. Component Ontologies

Our framework combines and extends two existing components: The workflow model and biomedical ontology. The workflow model consists of a process model and an organizational (participants/role) model. The process model can represent ordering of processes (e.g., protein translation) and the structural components that participate in them (e.g., protein). Processes may be of low granularity (high-level processes) or of high granularity (low-level processes). High-level processes are nested to control the complexity of the presentation for human inspection. The participants/role model represents the relationships among participants (e.g., an EF-tu is a member of the elongation factors collection in prokaryotes) and the roles that participants play in the modeled processes (e.g., EF-tu has enzymic function: GTPase). We used the workflow model as a biological process model by mapping workflow activities to biological processes, organizational units to biomolecular complexes, humans (individuals) to their biopolymers and networks of events, and roles to biological processes and functions.

A significant advantage of the workflow model is that it can map to Petri Nets, a mathematical model that represents concurrent systems, which allows verification of formal properties as well as qualitative simulation. A Petri Net is represented by a directed, bipartite graph in which nodes are either places or transitions, where places represent conditions (e.g., parasite in the bloodstream) and transitions represent activities (e.g., invasion of host erythrocytes). Tokens that are placed on places define the state of the Petri Net (marking). A token that resides in a place signifies that the condition that the place represents is true. A Petri Net can be executed in the following way. When all the places with arcs to a transition have a token, the transition is enabled, and may fire, by removing a token from each input place and adding a token to each place pointed to by the transition. High-level Petri Nets, used in this work, include extensions that allow modeling of time, data, and hierarchies.

For the biomedical ontology, we combine the Transparent Access to Multiple Biological Information Sources (TAMBIS) with the Unified Medical Language System (UMLS). TAMBIS is an ontology for describing data to be obtained from bioinformatics sources. It describes biological entities at the molecular level. UMLS describes clinical and medical entities. It is a publicly available federation of biomedical controlled terminologies and includes a semantic network with 134 semantic types that provides a consistent categorization of thousands of biomedical concepts. The 2002AA edition of the UMLS Metathesaurus includes 776,940 concepts and 2.1 million concept names in over 60 different biomedical source vocabularies. We augmented these two core terminological models to represent mutations and their effects on biomolecular structures, biochemical functions, cellular processes, and clinical phenotypes. The extensions include classes for representing (1) mutations and alleles and their relationship to sequence components, (2) a nucleic acid three-dimensional structure linked to secondary and primary structural blocks, and (3) a set of composition operators, based on the nomenclature of composition relationships, due to Odell.

Odell introduced a nomenclature of six kinds of composition. We are using three of these composition relationships in our model. The relationship between a biomolecular complex (e.g., ternary complex) and its parts (e.g., GTP, EF-tu, aa-tRNA) is a component–integral object composition. This relationship defines a configuration of parts within a whole. A configuration requires the parts to bear a particular functional or structural relationship to one another as well as to the object they constitute. The relationship between an individual molecule (e.g., tRNA) and its domains (e.g., D domain, T domain) is a place–area composition. This relationship defines a configuration of parts, where parts are the same kind of thing as the whole and the parts cannot be separated from the whole. Member–bunch composition groups together molecules into collections when the collection members share similar functionality (e.g., elongation factors) or cellular location (e.g., membrane proteins). We have not found the other three composition relationships due to Odell to be relevant for our model.

We implemented our framework using the Protégé-2000 knowledge-modeling tool. We used Protégé's axiom language (PAL) to define queries in a subset of first-order predicate logic written in the Knowledge Interchange Format syntax. The queries present, in tabular format, relationships among processes and structural components as well as the relationship between a defective process or clinical phenotype and the mutation that is causing it.

1.2.2. Translation into Petri Nets

We manually translated the tRNA workflow model into corresponding Petri Nets, according to mapping defined by others. The Petri Net models that we used were high-level Petri Nets that allow the representation of hierarchy and data. Hierarchies enable expanding a transition in a given Petri Net to an entire Petri Net, as is done in expanding workflow high-level processes into a net of lower level processes. We upgraded the derived Petri Nets to Colored Petri Nets (CPNs) by:

1. Defining color sets for tRNA molecules (mutated and normal), mRNA molecules, and nucleotides that comprise the mRNA sequence and initiating the Petri Nets with an initial marking of colored tokens

2. Adding guards on transitions that relate to different types of tRNA molecules (e.g., fMet-tRNA vs. elongating tRNA molecules)

3. Defining mRNA sequences that serve as the template for translation

We used the Woflan Petri Net verification tool to verify that the Petri Nets are bounded (i.e., no accumulation of an infinite amount of tokens) and live (i.e., deadlocks do not exist). To accommodate limitations in the Woflan tool, which does not support colored Petri Nets, we manually made several minor changes to the Petri Nets before verifying them. We simulated the Petri Nets to study the dynamic aspects of the translation process using the Design CPN tool, which has since been replaced by CPN Tools.

1.2.3. Sources of Biological Data

We gathered information from databases and published literature in order to develop the tRNA example considered in this work. We identified data sources with information pertaining to tRNA sequence, structure, modifications, mutations, and disease associations. The databases that we used were:

• Compilation of mammalian mitochondrial tRNA genes, aimed at defining typical as well as consensus primary and secondary structural features of mammalian mitochondrial tRNAs (http://mamit-trna.u-strasbg.fr/)

• Compilation of tRNA sequences and sequences of tRNA genes (http:// www.uni-bayreuth.de/departments/biochemie/sprinzl/trna/)

• The Comparative RNA website (http://www.rna.icmb.utexas.edu/), which provides a modeling environment for sequence and secondary-structure comparisons

• Structural Classifications of RNA (SCOR, http://scor.lbl.gov/scor.html)

• The RNA Modification Database (http://medlib.med.utah.edu/RNAmods), which provides literature and data on nucleotide modifications in RNA

• A database on tRNA genes and molecules in mitochondria and photosynthetic eukaryotes (http://www.ba.itb.cnr.it/PLMItRNA/)

• Online Mendelian Inheritance in Man (OMIM) (http://www.ncbi.nlm.nih. gov/omim/), which catalogs human genes and genetic disorders

• BioCyc (http://metacyc.org/), a collection of genome and metabolic pathway databases which describes pathways, reactions, and enzymes of a variety of organisms

• Entrez, the life sciences search engine, which provides views for a variety of genomes, complete chromosomes, contiged sequence maps, and integrated genetic and physical maps (http://www.ncbi.nlm.nih.gov/gquery/ gquery.fcgi?itool = toolbar)

• MITOMAP, A human mitochondrial genome database (http://www. mitomap.org/)

• The UniProt/Swiss-Prot Protein Knowledgebase, which gives access to wealthy annotations and publicly available resources of protein information (http://us.expasy.org/sprot/sprot-top.html)

In addition, we used microarrays and mass spectral data, providing information on proteins involved in tRNA processing or affected by tRNA mutations.

1.3. MODELING APPROACH AND RESULTS

Our model represents data using process diagrams and participant/role diagrams. Appendix A on our website (http://mis.hevra.haifa.ac.il/~morpeleg/NewProcess Model/Malaria_PN_Example_Files.html) presents the number of processes, participants, roles, and links that we used in our model. The most granular thing that we represented was at the level of a single nucleotide (e.g., GTP). The biggest molecule that we represented was the ribosome. We chose our levels of granularity in a way that considers the translation process under the assumption of a perfect ribosome; we only considered errors in translation that are due to tRNA. This assumption also influenced our design of the translation process model. This design follows individual tRNA molecules throughout the translation process and therefore represents the translocation of tRNA molecules from the P to the E site and from the A to the P site as distinct processes that occur in parallel. The level of detail in which we represented the model led us to consider questions such as (1) "Can tRNA bind the A site before previously bound tRNA molecule is released from the E site?" and (2) "Can fMet tRNA form a ternary complex?"
(Continues...) Excerpted from Genomics and Proteomics Engineering in Medicine and Biology by Metin Akay. Copyright © 2007 by Institute of Electrical and Electronics Engineers, Inc.. Excerpted by permission of John Wiley & Sons.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.

Current applications and recent advances in genomics and proteomics

Genomics and Proteomics Engineering in Medicine and Biology presents a well-rounded, interdisciplinary discussion of a topic that is at the cutting edge of both molecular biology and bioengineering. Compiling contributions by established experts, this book highlights up-to-date applications of biomedical informatics, as well as advancements in genomics-proteomics areas. Structures and algorithms are used to analyze genomic data and develop computational solutions for pathological understanding.

Topics discussed include:

• Qualitative knowledge models
• Interpreting micro-array data
• Gene regulation bioinformatics
• Methods to analyze micro-array
• Cancer behavior and radiation therapy
• Error-control codes and the genome
• Complex life science multi-database queries
• Computational protein analysis
• Tumor and tumor suppressor proteins interactions