Pathway/Genome Database Concepts Guide
4 PGDB Reactions
6 PGDB Pathways
This document describes concepts involved in Pathway/Genome Databases (PGDBs) managed by the Pathway Tools software, such as those in the BioCyc PGDB collection.
PGDB data is accessible in several ways, which are described in more detail on the downloads page.
As DNA sequencing has become very affordable, many fairly similar strains have been sequenced for some popular organisms of research interest, leading to an explosion of mostly similar data, but hopefully containing some interesting differences somewhere. A Pan-Genome PGDB collects similar strains into one combined PGDB, to highlight the core genomic essence of a species, and to show what fringe parts of the strain genomes are undergoing active evolution and diversification. The core, shared genes are determined by orthology.
The following steps are taken to construct a Pan-Genome PGDB:
When viewing the Cellular Overview for a Pan-Genome PGDB, two special highlighting commands are made available. Highlighting the Core Genes shows all the reactions of the genes that are shared among all the strain PGDBs, in other words, each gene has orthologs to all the other strains. Highlighting the Unique Genes shows all the reactions of the genes that have no orthologs at all, and are thus uniquely contributed by only one single strain. There is an additional set of genes that are not shared by all strains, but at least some of them. Currently, there is no simple way offered yet to show these varied levels of shared genes.
This section introduces concepts relevant to PGDB reactions.
PGDBs can encompass multiple types of reactions, which are classified using the PGDB reaction ontology. Most “standard” metabolic reactions are instances of the class Chemical-Reactions, which is a child of class Simple-Reactions. The main classes in the reaction ontology are as follows.
How do PGDBs handle reaction direction?
In a PGDB, each reaction is stored as an object that is an instance of the Reactions class. That object includes two slots, Left and Right, each of which contains a list of chemicals that are the reactants and products (not necessarily respectively) of the reaction. That is, in some cases the Left slot stores the reaction reactants, in other cases the Left slot stores the products.
Following the conventions used by the IUBMB Enzyme Commission , the direction in which a reaction is stored in a PGDB has no implication for the physiological directionality of that reaction. In the IUBMB EC system, all reactions within a given class are written in a single consistent direction (e.g., all hydrolases are written in the hydrolysis direction). Reactions categorized by the IUBMB EC system are stored within a PGDB such that the EC left side of the reaction is stored in the Left slot.
The PGDB framework for defining reaction direction is designed both for flexibility in encoding a diverse range of biological situations, and to minimize the work curators must do to define reaction directions. The diverse biological situations to be encoded include the notions that the directionality of some reactions are invariant, whereas the directionalities of other reactions depend on the enzyme that is catalyzing the reaction, and on the organism in which the reaction occurs.
The directionality of some reactions is explicitly stored within the PGDB. The directionality of other reactions is not stored, but is computed on demand by Pathway Tools. The best way to query the directionality of a reaction is via the slot Reaction-Direction in reaction objects. Even when no value is explicitly stored in this slot, a method attached to this slot will attempt to compute a value for the slot. Possible values of this slot are as follows. The values PHYSIOL-LEFT-TO-RIGHT, IRREVERSIBLE-LEFT-TO-RIGHT, and LEFT-TO-RIGHT mean that the Left slot should be treated as containing the reactants; the values PHYSIOL-RIGHT-TO-LEFT, IRREVERSIBLE-RIGHT-TO-LEFT, and RIGHT-TO-LEFT mean that the Right slot contains the reactants.
The software computes values of the Reaction-Direction slot by integrating information within the enzymes and pathways associated with the reaction. Consider these examples where no information is stored in the Reaction-Direction slot of reaction R:
In general, if the software finds information indicating that R proceeds in both the left-to-right and the right-to-left directions, then it infers that R is reversible.
The equilibrium constant and the change in Gibbs free energy stored for the reaction (if any) refer to the direction of the reaction as stored, that is, assuming that the Left slot contains the reactants.
The Enzyme Commission (EC) classifies enzymes based on the reactions they catalyze. In addition to a description of the enzymatic activity, each classified enzyme receives a descriptive and accurate name and a unique number, known as an EC number. The use of EC numbers makes it possible for scientists to refer to enzymes in a consistent and unambiguous way. In addition, by annotating genes with EC numbers, it is possible to computationally link those genes to precise enzymatic activities.
MetaCyc contains four types of EC numbers. The following list explains the differences among the different types.
An accurate representation of metabolic processes must take into account the existence of multiple cellular compartments and structures, and that the processes and metabolites are partitioned among these compartments. Even a simple cell consisting of a single compartment has a membrane boundary to the extracellular space, which is crossed by transport reactions. This section describes how Pathway Tools records the compartments where reactions and metabolites occur.
In Pathway Tools, compartments are specified using the controlled vocabulary in the CCO (Cell Component Ontology). Every CCO term is represented by a PGDB frame that is a child of the class CCO-SPACE or of the class CCO-MEMBRANE. At the time a PGDB is created, we infer the set of CCO terms applicable to the organism based on its taxonomic class (for example, eukaryotes will have all the terms associated with the cell nucleus, whereas prokaryotes will not — curators can edit the set of cell components present in an organism if it differs from the inferred default) and create instance frames for each such CCO class. In general the ids for CCO classes start with CCO-, whereas the ids for instances of CCO classes start with CCI-. In PGDBs with a single cellular architecture, it makes little effective difference whether a reaction location refers to a CCO instance frame or its parent class, but the reaction editor will not allow curators to assign reactions to non-abstract (see below) CCO classes that do not have any instances (because those components are not present in cells for that organism).
The default compartment for metabolites and reactions is assumed to be CCO-CYTOSOL in the S case, if no other information is specified in the reaction’s RXN-LOCATIONS slot that would override this default. In the T case, nothing can be assumed about the default locations, if no membrane and other mappings for CCO-IN and CCO-OUT were specified (see below). Such T reactions could not usefully become part of a reaction network model.
To avoid unnecessarily duplicating information, we store frames for metabolites and reactions once within each PGDB, even if they may simultaneously be present in more than one compartment in a PGDB. Instead, the compartments are specified by auxiliary information attached to reactions.
There are two types of reactions:
By default (meaning if no substrate locations are specified), S reactions are assumed to occur in the cytosol (CCO-CYTOSOL). To store non-default compartment information, reactions have a slot called RXN-LOCATIONS, the values of which differ between the S and T type reactions, as follows:
Whenever a reaction is transferred between PGDBs (by import or schema upgrade operations), all values in the RXN-LOCATIONS are filtered away (i.e. not copied). This prevents inapplicable compartments from being introduced into other PGDBs. In the future, PathoLogic and TIP will infer compartment information. The values of the RXN-LOCATIONS are listed in the flat files of the PGDB.
If the reaction is catalyzed by more than one enzyme (i.e. it has more than one enzymatic-reaction attached), then each value in the RXN-LOCATIONS slot must have an annotation called ENZRXNS, which has as its values the frame IDs of the corresponding enzymatic-reactions. This approach allows determining the precise compartment(s) in which the catalyzed reaction is occurring.
This section addresses the state of reaction mass balance and protonation state of chemical compounds in PGDBs. Because these issues are still evolving and are influenced to a large degree by history, we include a historical discussion of these issues.
Our long-term goal is for all PGDB reactions to be fully mass balanced and charge balanced, and for all chemical compounds to be properly protonated at cellular pH. Although in some cases such a treatment may yield reactions or chemical structures that look non-traditional to biochemists, we believe this approach provides the most consistent and correct treatment. In addition, it provides a treatment that will facilitate automatic generation of flux-balance models from PGDBs.
Historically, the chemical structure data within BioCyc databases has been obtained from many different sources, including textbooks, articles from the primary research literature, and downloading from certain open databases. In the early years of the project we developed programs to check the mass balance and element balance of reactions within PGDBs. We found that these programs were extremely valuable because identification of unbalanced reactions allowed us to identify errors in both the reaction equations, and in the chemical structures. However, we also found that, because of the diverse sources from which we obtained chemical structure data, the structures were protonated inconsistently. Therefore, for many years we ignored element imbalances due to hydrogen only, while correcting imbalances due to other elements.
In 2008, we began to address the problem of inconsistent protonation to facilitate automatic generation of flux-balance models by automatically protonating unbalanced reactions. The first release of the newly mass-balanced MetaCyc and EcoCyc DBs was version 13.0 in early 2009. In 2015, improved software tools for checking charge balance were used to obtain fully mass- and charge-balanced reactions in MetaCyc. In time, other BioCyc PGDBs will become fully mass- and charge-balanced as well as they are regenerated from newer versions of MetaCyc.
The following sections describe the methodology by which the protonation-state normalization and reaction mass balancing were achieved.
For a given chemical compound, there can be atoms that will bind a variable number of hydrogen atoms, depending on their chemical structure and the pH of their environment. A term for the isomers of a compound that differ in the number of hydrogens bound to these atoms is proto-isomer. A term for the atoms with variable numbers of bonded hydrogens is the proto-isomerization centers of a compound. Oxygen, sulfur, phosphorus, and nitrogen are examples of typical proto-isomerization centers.
In order to bring a greater degree of consistency to our PGDBs, we protonated (i.e., assigned the correct number of bound hydrogens to the proto-isomerization centers of a compound) the compounds of EcoCyc with a reference pH value of 7.3, using the Marvin (version 5.1.02) computational chemistry software available from ChemAxon, Ltd . The pH value of 7.3 was selected based on a paper on the measurement of cytoplasmic pH of E. coli . In order to easily exchange compound data between MetaCyc and EcoCyc, MetaCyc was also protonated with a reference pH value of 7.3. This step is an approximation since MetaCyc contains reactions and compounds from many organisms and many cellular compartments.
The Marvin software calculates the protonation state of a compound’s proto-isomerization centers by first determining their pKA. The pKA of the proto-isomerization centers of a compound were obtained by computing the partial charge distribution. This, in turn, is calculated using a numerical partial differential equation solver, which computes the distribution by means of the structure of the compound, and the known electronegativities of the constituent atoms. Although we have worked with ChemAxon to improve the accuracy of their calculations to match that of experimentally-verified pKA’s of many biochemically-relevant compounds, this calculation is still based on an approximation technique, and will not necessarily yield fully correct pKA’s for every substance.
Some caveats about our protonation of compounds:
Once the compounds of EcoCyc and MetaCyc were protonated, all reactions that had a mass-imbalance were identified and either balanced or labeled as unbalanceable. Some caveats about our computational reaction balancing:
This table provides information on the small-molecule reaction balance state for both EcoCyc and MetaCyc at different points in time. The categories below represent reactions that are balanced, unbalanced, and those for which it is not possible to determine the balance state.
For the category of reactions where it is not possible to determine the balance state, these are mainly due to:
The atom mapping of a reaction describes for each non-hydrogen atom in a reactant compound its corresponding atom in a product compound. The bonds broken and made by a reaction can be inferred from an atom mapping of that reaction. On the reaction page this correspondence is shown using matching colors for atoms and bonds, between reactants and products, or/and numbers labeling the atoms. If a bond is broken or made by the reaction, the bond is black. Some reactions have more than one atom mapping. In some cases not all mappings are not biologically relevant (e.g., the enzyme never produces them). It might also be the case that some of these multiple atom mappings are essentially the same due to symmetries within compound structures, or the inability to distinguish among some atoms. For some reactions no atom mapping are shown since none were computed for them. This lack of an atom mapping might result from several possible factors including to the complexity of the reaction (e.g., large substrates), that the reaction operates on substrates that have no structures, or the reaction is not mass balanced.
When Pathway Tools displays a reaction, it first tries to obtain an atom mapping from that reaction in its PGDB. However, typically reactions in PGDBs other than MetaCyc do not contain atom mappings (except for reactions not found in MetaCyc). Thus, the software next tries to find the same reaction in MetaCyc and use its atom mapping, if any, for that reaction.
The atom mappings in MetaCyc were computationally predicted without manual curation, but we expect a very low rate (< 3%) of errors. The approach used to compute these atom mappings was published in . Essentially, this approach computes atom mappings that minimize the overall cost of bonds broken and made in the reaction, given assigned propensities for bond creation and breakage.
Atom mapping data are available in three ways. Note: some reactions have more than one atom mapping.
As an example, the encoding of one atom mapping for reaction R524-RXN is:
REACTION - R524-RXN NTH-ATOM-MAPPING - 1 MAPPING-TYPE - NO-HYDROGEN-ENCODING FROM-SIDE - (HCO3 0 3) (CPD-69 4 6) TO-SIDE - (CARBAMATE 0 3) (CARBON-DIOXIDE 4 6) INDICES - 4 5 6 3 0 2 1 //
Each atom mapping of a reaction is encoded using six fields:
Before discussing this specific example, notice that the direction of the reaction in the PGDB (as given by the PGDB slots Left and Right) is not made use of in atom mappings. In particular, the From-Side in the mapping could be the left or the right side of the PGDB reaction. Also, the order of the compounds given by the From-Side and To-Side, and as discussed below, might not be the same as the order given by the Left side or Right side of the reaction. Reconstructing the atom mapping relative to the reaction must be performed relative to the chemical structures of the compounds, not the encoding of the reaction.
For this example we consider reaction R524-RXN (which is the unique id (i.e. frame id) of the reaction in MetaCyc). That reaction has only one atom mapping, which is identified as NTH-ATOM-MAPPING 1. The mapping type is NO-HYDROGEN-ENCODING. This mapping type tells us that the hydrogen atoms are not mapped. The From-Side compounds are bicarbonate (frame id HCO3) and cyanate (CPD-69), and the To-Side compounds are carbamate and carbon dioxide.
In this atom mapping only the non-hydrogen atoms are indexed. For the From-Side, HCO3 has 4 atoms mapped; they are indexed 0 to 3 (its one hydrogen atom is not indexed). CPD-69 has 3 atoms mapped; they are indexed 4 to 6. For the To-Side, CARBAMATE has 4 atoms mapped that are indexed 0 to 3. CARBON-DIOXIDE has three atoms mapped that are indexed 4 to 6. Note that the indices of the compounds form a continuous sequence on each side, which means that the indices of the atoms within these compounds are shifted accordingly. The example below shows this aspect more precisely.
The permutation indices (INDICES) is the final component in describing the mapping of atoms from the From-Side to the To-Side. An index integer j located at position i of INDICES gives the mapping of atom j of the From-Side to the corresponding atom i of the To-Side. For the R524-RXN example above, we have the following atom mapping (note that the index values start at 0). The first number in indices is 4 (j = 4), and it is in position 0 in the list of indices (i = 0). Since j refers to the From-Side, the 4 identifies the atom with index 4 on the From-Side. The FROM-SIDE data tells us that HCO3 spans atoms 0–3, and CPD-69 spans atoms 4–6, thus j = 4 identifies atom 4 overall in the list that spans both compounds, which is the 0th atom of CPD-69. Similarly, i = 0 identifies the 0th atom of the To-Side. The TO-SIDE data tells us that CARBAMATE spans atoms 0–3 and CARBON-DIOXIDE spans atoms 4–6, thus i = 0 refers to the 0th atom in the overall list, which is also the 0th atom of CARBAMATE. When we say in the first line below that “4 is atom 0 (C) of cyanate,” we are translating from the indexing system for the overall From-Side (4th atom in the overall list) to the indexing system for cyanate alone (0th atom within the chemical structure for cyanate).
The chemical structures of the compounds (and the numbering of atoms for each compound) are not encoded within the atom mapping. They are stored in the PGDB objects that describe each compound, and are available in several forms:
The compound-specific atom indices used in atom mappings refer to the index of each atom in the chemical structure as stored in the PGDB, such as in the atom section of a MOL file.
The SMILES syntax allows not only the representation of chemical structures and of reactions, but also representation of reaction atom mappings. The full description of the SMILES syntax is given at the following external link: SMILES Tutorial. We will only give a summary of that syntax and one example.
A reaction such as THIOSULFATE-SULFURTRANSFERASE-RXN is: described by the equation
thiosulfate + hydrogen cyanide => sulfite + thiocyanate + 2 H+
has one atom mapping in MetaCyc. Using SMILES, this atom mapping is represented in the following way:
The first part on the left before the dot, namely
The atom mapping is encoded by using integer labels between square
brackets. For example,
As mentioned in Section 4.6.1, atom mapping data based on SMILES can be obtained by downloading the file atom-mappings-smiles.dat from the MetaCyc download bundle. That file contains one reaction per line, the first element on each line is the frame id of the reaction, followed by a tab, then one or more SMILES separated by a space, one for each atom mapping of that reaction.
The atom mapping encoding presented in the previous subsection is the result of a canonicalization (i.e., a normalization) process. You do not need to know how this process is working to use and decode the atom mappings but it might be useful if you want to create your own atom mapping encoding implementation that will result with the same atom mapping as the ones in your PGDB.
Canonicalization has one major goal: The atom mapping encoding does not depend on the manner in which the compounds and reactions are stored in a PGDB.
The encoding was also designed such that the atom mapping can be reconstructed from the compound structures without referring to the reaction.
The encoding depends on the compound InChi strings and the ordering of their atoms given by the program computing the InChi string. The process of canonicalization is as follows:
The computation (i.e., estimation) of the standard Δ Gibbs free energy for reactions and compounds in MetaCyc, that is Δr G⁄○ and Δf G⁄○, respectively, was done at pH 7.3 and ionic strength 0.25. We used pH 7.3 because the computation of the protonation state of all compounds in MetaCyc used that value. The computation of the standard Gibbs free energy of change formation of compounds is first done by an estimation at pH 0 and ionic strength 0 (Δf G○) based on the technique presented in . This technique is based on the decomposition of the compounds into known “contribution groups”. Then, the standard Gibbs free energy at pH 7.3 and ionic strength 0.25 (Δf G⁄○) is computed based on a technique developed by Robert A. Alberty . In his technique, Alberty proposes to use several protonation states for some compounds, but we had to simplify it by always using only one protonation state by compound, that is, the unique one stored in MetaCyc.
For the standard Gibbs free energy of reactions, Δr G⁄○, the computation is based on the Δf G⁄○ values of the compounds involved in the reaction. The Δf G⁄○ could not be computed for some of the compounds in MetaCyc due to the impossibility to decompose them into the contribution groups provided by the technique of . Consequently, the Δr G⁄○ is not computed for any reaction which has a substrate for which its Δf G⁄○ is not stored in MetaCyc.
This section introduces concepts relevant to PGDB metabolic and signaling pathways. Many of the issues discussed here are also explored in .
Pathway boundaries  are defined heuristically, using the judgment of expert curators. More information about curation practices is available in the Curator Guide. Curators consider the following aspects of a pathway when defining its boundaries.
The preceding philosophy toward pathway boundary definition contrasts sharply with KEGG maps. KEGG maps are on average 4.2 times larger than MetaCyc pathways because KEGG tends to group into a single map multiple biological pathways that converge on a single metabolite [Pathway05] .
We define a super-pathway as a cluster of related pathways. Typically, a super-pathway consists of a linked set of smaller pathways that share a common metabolite. For example, the super pathway superpathway of phenylalanine, tyrosine, and tryptophan biosynthesis consists of several pathways that converge at the metabolite chorismate. The components of super-pathways include base pathways (pathways that are not themselves super-pathways), other super-pathways, and individual reactions that have not necessarily been assigned to base pathways. Those reactions typically serve to connect together the component pathways within a super-pathway.
Super-pathways are stored within each PGDB – they are not computed dynamically.
Experimentalists have elucidated variants of a given metabolic pathway in different organisms. These pathway variants  are defined as distinct pathway objects in MetaCyc when they differ in their constituent reactions. For example, MetaCyc contains several variations of the TCA cycle that have been observed in different organisms. Variant pathways are assigned names such as “TCA cycle variation II” and “arginine degradation III” (meaning the third form of arginine degradation in MetaCyc). Two pathways are not considered variants of one another in MetaCyc if their constituent reactions are identical. For example, if the glycolysis pathway occurs with identical constituent reactions in two different organisms, but with different enzymes (different protein sequences), these occurrences of glycolysis are not considered to be pathway variants. When copied to other Pathway/Genome databases [def] by the PathoLogic pathway prediction program, the names of variant pathways are not changed. This approach provides for consistent naming across DBs, but does have the side effect that names may seem inconsistent if considered only in the context of one DB, such as a DB that contains only “arginine degradation III” and “arginine degradation VIII”.
We define  a conspecific pathway as a pathway that belongs to a specific species, whereas a chimeric pathway is a pathway that combines elements from many species but does not occur in any one species in its entirety. MetaCyc contains both types of pathways, which are labeled as such.
No. Pathways comprise a layer defined on top of the reaction-based metabolic network. Users can choose to compute with the metabolic (reaction) network directly, ignoring the pathway layer, if they so choose. Note that every PGDB contains some metabolic reactions that are not assigned to any metabolic pathway.
PGDB objects contain evidence codes that describe the types of evidence that support and justify the inclusion of that entry in the PGDB. For example, evidence codes on an enzyme entry indicate the type of evidence supporting the inference that the enzyme catalyzes its associated biochemical reaction. See  for a general description of the evidence-code system. The evidence-code ontology is available here.
Because the data in MetaCyc are derived from the experimental literature, the vast majority of evidence codes within MetaCyc will be experimental evidence codes that identify different classes of experimental methods that support data within MetaCyc. However, the evidence-system supports a larger set of evidence types, including evidence based on computational inferences, which are used more extensively in PGDBs other than MetaCyc.
Evidence codes appear as icons in the upper-right corner of displays pages such as pathway and enzyme pages (example). For example, a flask icon on a metabolic pathway page indicates that experimental evidence supports the existence of that pathway. Often a detailed evidence code has been assigned that indicates a specific type of experimental method – click on the flask icon to see a description of the type(s) of experimental method(s) used to elucidate the pathway (when available), and for associated citations (when available).
Gene Ontology (GO) terms from all three GO taxonomies (biological process, molecular function, cellular component) can be associated with either polypeptides or protein complexes. Ideally, GO terms that refer to the activity of a complex should be assigned to the complex directly rather than to its component gene products. However, since GO terms imported from external sources are often associated only with gene products, either is acceptable. Each GO term is represented by a class object that includes its description, links to its parent and child terms via both the is-a and part-of relationships, and links to all the objects in the PGDB that are directly annotated with that term. Each PGDB contains objects only for those GO terms actually used in the PGDB, i.e. those to which proteins are annotated plus all ancestors of those terms. The complete GO taxonomy is maintained in a separate database and updated periodically – terms from this database are imported into a PGDB if necessary in the course of annotating a protein. Pre-release consistency checks ensure that the GO terms in a PGDB match those in the master database, and any proteins annotated to obsolete terms are flagged for attention from a curator.
The assignment of a GO term to a protein should be annotated with an evidence code and citation. The evidence code is one from the Pathway Tools evidence code ontology described above, but a straightforward mapping between Pathway Tools evidence codes and Gene Ontology evidence codes allows easy conversion from one to the other. The evidence code assignment may include the GO with/from field (although it is not required and there are no restrictions on its use or format). No other GO qualifiers (e.g. not, contributes-to) are supported.
The GO terms for a protein are listed with their evidence codes and citations in a table on the protein page. Clicking on a GO term shows the page for that term, including links to all proteins annotated with that term. GO terms can be supplied as search criteria on the Gene/Protein/RNA Search page. In addition, the cellular component GO terms serve as the repository of protein cellular location information in Pathway Tools. A mapping between cellular component GO terms and Pathway Tools CCO terms (described in Section 4.4) allows the protein LOCATIONS slot (which describes the protein location using CCO terms) to be computed directly from the GO terms.
We have developed a number of procedures and software tools to ensure the quality of PGDB data.
The interactive editing tools that curators use to update PGDBs implement a number of quality checks and warn curators when entered data does not pass these checks. For example, the reaction editor warns the user if an entered reaction is not balanced. The editors warn the user if required fields have not been entered. They ensure that entered data is of the proper type, such as allowing the user to choose from a controlled vocabulary of choices, or ensuring that specific fields are filled with objects of a specific type (e.g., the entities listed as reactants and products of a reaction must be chemical compounds). Syntax checking is also performed, such as checking that any HTML markup within mini-review text is properly formatted.
The Consistency Checker is a large set of data checking tools that are run individually or as a group on a given PGDB. For example, they are run by curators shortly before each database release, on every PGDB that has been manually updated. Some of the tools identify data quality problems and report them so that curators can manually fix the problems. Other tools both find problems and automatically repair them.
The Consistency Checker includes the following tools.
The Consistency Checker also includes tools that compute derived information for a PGDB, such as the following. These tools are run on both curated and uncurated PGDBs.
A special set of checks are performed on newly generated PGDBs. Curators examine sample PGDBs manually to look for incorrect inferences or other problems. In addition we compute statistics across newly generated PGDBs and discard PGDBs that do not contain at least 300 genes and 10 pathways — such genomes usually have such bad annotations that their data is of little value.