Metabolic Modeling to Predict Organism Phenotypes

Here we explore one of the major applications of steady-state metabolic modeling: the prediction of organism growth rates under varying perturbations.  The two most common perturbations studied with metabolic models are variations in the nutrients available to the organism (e.g., changes in carbon source, nitrogen source, and oxygen availability), and the presence of gene knockouts.  These two perturbations can be combined since the effects of gene knockouts can be modeled under different nutrient mixes. 

Metabolic Modeling for Validation of Genome Annotations

A major advance in bioinformatics in the last decade is the rapidity with which we can now create quantitative metabolic models from sequenced genomes.  In this and future blog posts we will examine several applications of metabolic modeling.  This post introduces metabolic modeling, considers its use for validation of genome annotations, and proposes that construction of metabolic models can form a routine part of the genome annotation process.

Searching for Metabolic Routes in Pathway Tools

The Metabolic Route Search Problem

Consider the problem of performing an in-depth exploration of the metabolic network of an organism that you study, to compare alternative paths within that network whereby the organism can transform a starting metabolite into an ending metabolite.  What are the lengths and properties of these alternative pathways? 

Consider now a broader problem, namely the metabolic-engineering problem of finding the most efficient modification to the biochemical network of an organism to allow the organism to synthesize a new metabolite from a feedstock compound.  One aspect of "most efficient" is minimize the number of reactions added from an external database of known reactions.

RouteSearch [1] is a Pathway Tools component that solves both of the preceding problems by computing optimal metabolic routes, that is, an optimal series of biochemical reactions that connects start and goal compounds, given various cost parameters to control the optimality of the routes found.  RouteSearch can display several of the best routes it finds using an interactive graphical web page.  When RouteSearch is used for metabolic engineering, it uses the MetaCyc database as its external reaction database.

In computing optimality, RouteSearch takes into account the conservation of nonhydrogen atoms from the start compound to the goal compound. Perhaps surprisingly, it is possible to devise reaction paths that conserve no atoms from start to goal compound!  The more atoms that are conserved, the more efficient the transformation from start to goal.  To compute the number of conserved atoms, RouteSearch uses precomputed atom mappings of reactions that are available in MetaCyc [2]. An atom mapping of a reaction gives a one to one correspondence of each nonhydrogen atom from reactants to products.

RouteSearch is available only in Web mode in Pathway Tools (since version 17.0, March 2013). It is also available at BioCyc.org but without the possibility to add reactions from MetaCyc (that mode is available only for locally installed versions of Pathway Tools). More details on how to use MetaCyc with RouteSearch are given in the following section.

Propagating Updates from MetaCyc -- Nearly Effortless Improvements to your PGDB!

New versions of Pathway Tools are released every six months, with the current version being 18.5. Included in each new release is an updated version of the MetaCyc database. The curators at SRI are constantly working to improve MetaCyc, both adding new information, and fixing errors in the existing data. Here are some of the kinds of changes you can expect to see with each new release:

• New compounds, reactions, enzymes and pathways
• Addition of compound structures to existing compounds that previously lacked them
• Fixes to errors and other improvements to existing compound structures
• Updates to reaction equations to fix errors and so that they balance and are correctly protonated
• Updates to pathways to fix errors and incorporate new knowledge
• Updates and addition of EC Numbers, names, literature citations, links to other databases, and textual summaries.

Any new PGDB you create will have its reactions, pathways and metabolites imported from the most recent version of MetaCyc and therefore benefit from all the latest changes.  But what about your old PGDBs? If they were created with an older version of MetaCyc, then more and more of their information will become outdated over time. When you open an old PGDB in a new version of Pathway Tools, you will be asked to upgrade it. Upgrading applies schema changes that are necessary in order for the PGDB to be able to operate with the new software, but it will not incorporate the MetaCyc data updates. To incorporate MetaCyc updates into your PGDB, you must invoke the Tools → Propagate MetaCyc Data Updates command for each of your PGDBs.

Why should I use this tool?

Aside from the obvious benefits of fixing errors in your PGDB, there are several reasons why it is a good idea to propagate MetaCyc updates with every new release.

• It will allow more of your reactions to balance, improving atom mapping and Route Search computations.
• It will greatly improve the process and results of building metabolic models via flux-balance analysis -- unbalanced reactions are automatically excluded from metabolic models by MetaFlux.
• It will facilitate apples-to-apples comparisons with other PGDBs built with (or updated to) the latest version of MetaCyc.

Metagenomics, PathoLogic and Pathway Abundance

Pathway abundance is a new parameter computed by PathoLogic, the tool that generates PGDBs in Pathway Tools. It is available starting with version 17.0 (March 2013) of Pathway Tools. Pathway abundances are computed from gene abundances supplied in metagenomics datasets, and are useful for comparing the metabolic profiles of different microbial communities.

Gene abundances are specified in the annotated genome file. Only the PathoLogic file format supports the specification of gene abundances. That is, gene abundance specification is not supported for the Genbank format. See Section “The PathoLogic File Format” in the Pathway Tools User Guide for more information about the PathoLogic format and how to specify the abundance attribute for a gene.

No preprocessing of the gene abundances is done by PathoLogic. That is, all gene abundances are taken as specified without doing any filtering such as outlier removal. We assume that any preprocessing of gene abundances has been done prior to generation of the annotated genome file. In particular, if some gene abundances are considered too low to be considered for the pathway abundances, these gene abundances should be omitted.

The abundance of a pathway is computed based on the gene abundances involved in the pathway. More precisely, assume that R is the set of reactions in pathway P for which gene abundances are specified, |R| the size of R and g_a is the given abundance of gene g. The abundance of pathway P is

  That is, the abundance of a pathway is the sum of the abundances of the genes catalyzing the reactions of the pathway, divided by the number of reactions of the pathway for which gene abundances are given. Notice that this formula does take into account all the known isozymes catalyzing a reaction and the spontaneous reactions do not take part in the computation.

Once PathoLogic has inferred the pathways from the annotated genome file, the computed abundances of the pathways can be found in the file pathways-report.txt under the subdirectory report of your PGDB. This report file lists all pathways that were inferred present in the PGDB alongside various computed parameters (e.g., confidence factor) including the computed abundances.

A Sequence Data File for Associating Enzymes with MetaCyc Reactions

Metabolic reconstruction typically proceeds in two steps: (1) analyze the proteome of a sequenced organism to infer the set of reactions catalyzed by the organism (the organism's reactome), and (2) infer the metabolic pathways present in the organism from the reactome.

Step (1) usually involves computing associations between protein sequences and the reactions in a pathway database such as MetaCyc. Such associations represent the inference that a given protein catalyzes that MetaCyc reaction.

Such associations can be inferred using a variety of sequence-analysis methods.  To aid researchers in associating sequences to MetaCyc reactions, each release of MetaCyc includes a file that associates MetaCyc reaction IDs with the UniProt identifiers of enzymes known to catalyze those reactions.  Note that not all MetaCyc reactions have EC numbers (because not all enzyme-catalyzed reactions have yet been assigned EC numbers), therefore EC numbers are not a comprehensive mechanism for associating sequences to reactions.  The file is called uniprot-seq-ids.dat and is included in the MetaCyc data file distribution.

The file contains a Lisp list of the form:

((RXN-1  EC#  ID1 ... IDn)
 (RXN-2  EC#  ID1 ... IDn)
 ...)

where RXN-1 is the MetaCyc unique ID of a MetaCyc reaction, EC# is the EC number of that reaction (or NIL if the reaction has not been assigned an EC number), and each ID1 etc are the UniProt IDs of UniProt proteins that catalyze RXN-1.

Note that when we prepare this file we intentionally filter out those UniProt proteins that have high sequence similarity to other UniProt proteins already listed for a given reaction, to bound both the size of the file and the cost of the downstream sequence comparisons.

Atom Mappings for Reactions

Starting with Pathway Tools version 16.5, atom mapping data is available for a large number of MetaCyc reactions. Atom mapping describes for each atom of a reactant (excluding hydrogen) its corresponding atom in the product. Implicitly, atom mapping illustrates which bonds are broken and created during the reaction. The reaction page displays atom mappings by using matching colors for atoms and bonds, between reactants and products, and/or numbers labelling the atoms. If a bond is broken or made by the reaction, the bond is black.  Atom mappings are shown on the desktop as well as in Web server mode, for Firefox and Chrome browsers.  Unfortunately, the support for SVG, used for Web reaction display, is not good enough in Internet Explorer and Safari, such that those browsers do not currently show the atom mappings.

The atom mappings were computed using the technique described in Latendresse et al., "Accurate atom-mapping computation for biochemical reactions", J. Chem. Inf. Model., September 2012.  A prerequisite for computing atom mappings is canonicalizing the order of atoms in each compound structure.  This means that an invariant atom order is determined that is independent of variations of how the structure is drawn.  InChi strings are used to perform this canonicalization.  All compounds in MetaCyc were canonicalized, and an InChi executable is bundled with the 16.5 distributions, to allow automatic canonicalization when curators create new compound structures with the Marvin editor.

While the atom mappings are stored in MetaCyc, they can also be displayed in another PGDB, such as EcoCyc, for a reaction identical to one in MetaCyc.  For this to work correctly, the compound structures in the other PGDB also have to be canonicalized, or else the MetaCyc atom mappings will not match.  We have propagated the canonicalized compounds to many of the Tier2 PGDBs of our BioCyc collection.  However, most of Tier3 and PGDBs created by outside users are unlikely to have been updated yet in this regard.

Users of Pathway Tools can fix this problem in the following way for a given PGDB.  Select the menu command Tools->Propagate MetaCyc Data Updates, which will bring up a dialog panel.  This operation, discussed in the User Guide, updates compound structures and reactions with the latest data from MetaCyc.  After this, display of atom mappings should work in the updated PGDB.

Please tell us how you like the atom mappings, and any interesting insights and uses of them that you find.  We would also be interested in reports where computed atom mappings do not in fact correspond to experimentally determined mappings, such that the computational method can be further improved.

FBA Tutorial Announcement

One of the exciting new features in the Pathway Tools 15.0 release is a Flux Balance Analysis (FBA) module.  FBA models are steady-state quantitative models of the metabolic network that can be used for predicting growth phenotypes under different nutrient conditions and under gene knock-outs.  An FBA model predicts the steady-state flux rates of metabolic reactions given a set of nutrients, secretions, and metabolites to produce (i.e. the biomass reaction), and the set of reactions in a PGDB.

There are two main tasks involved in flux balance analysis: developing a model, and then solving it.  Pathway Tools can aid in both of these tasks.  Developing an accurate FBA model might require the addition of new reactions to a PGDB, altering the directionality of some reactions, modifying the biomass reaction (which lists all the chemical components of the cellular biomass), and adding new nutrients and secretions. The Pathway Tools FBA module can guide the user by suggesting changes, such as possible new reactions to add to produce a needed compound.

Solving an FBA model involves determining the correct fluxes of reactions given a set of nutrients, secretions, and biomass metabolites to produce, and is performed once a feasible FBA model has been generated.  Pathway Tools uses the SCIP solver for this task.  Since the results of solving the model can point out problems in the original model, these two tasks, model-generation and solving, are often invoked in an iterative fashion. Results can also be viewed on the Cellular Overview Diagram.

We are pleased to announce that SRI will be hosting a tutorial on the new Pathway Tools module for generating and refining metabolic flux models using flux-balance analysis at SRI's main campus in Menlo Park, CA from June 21-22.  The purpose of the workshop is to describe the use of this new module in significant detail, to offer advice on strategies for refining FBA models, and to assist attendees in hands-on refinement of Pathway Tools based FBA models.

More details and registration are available at http://ptools-fba.eventbrite.com/  A remote online attendance option for the tutorial is available.

If you're at all interested in these new capabilities, please join us -- we'd love to see you!

Talkin' 'Bout My Regulation

The Regulation Summary Diagram for the bglG gene
in EcoCyc

You may have seen our new regulation summary diagrams on our EcoCyc gene pages. Or played with the regulatory overview. And you may have wished that you could get those visualizations for your own PGDBs. Unfortunately, there is as of yet no equivalent of PathoLogic for regulation -- no tool that will infer regulatory relationships, transcription factor sites, etc. from the genome annotation. Much of the regulatory data in EcoCyc was painstakingly curated from the primary literature (either by us or by the folks at RegulonDB) and entered piece by piece using our curation tools. That makes it an extremely valuable resource, but makes it difficult to replicate in other PGDBs without expending an equivalent amount of effort.

The Regulatory Overview Diagram for EcoCyc

Fortunately, there are other, faster ways to generate regulation data. High-throughput experiments and computational prediction programs can identify regulatory relationships and/or transcription factor binding sites en masse, and a number of groups have generated such data for their own organisms. The question that remains is how to bulk-load that data into a PGDB.

Are You Being Served?

A table of pathways with omics data.
(click to enlarge)

This post will provide a brief survey of the Pathway Tools Web Services.  These services can be used to access any site running the latest version (15.0) of Pathway Tools.  Thus, you can invoke them for any of the 1000+ PGDBs served by the BioCyc website without installing the Pathway Tools software locally, or you can install and run the software yourself in order to create queries (or allow other users to create queries) that access your own PGDBs.