Microbiome Metabolic Modeling with Pathway Tools

Transcription

1 Microbiome Metabolic Modeling with Pathway Tools Peter D. Karp SRI International

2 Cost for High Quality Models of 300 Microbiome Organisms Mul/ple body sites of human, mouse, farm animals 500 species in human pan metagenome (Nielsen) 321 organisms for human alone (MagnusdoJr) 300 models x 2 months/model x $100K/yr = $5M Assumes cost of combining models is zero If every model is duplicated 5 /mes: $25M

3 Challenges of Microbiome Metabolic Modeling Scale Possible approaches Increase speed of manual model building Collaborate / reuse organism models in community models Understandability Collabora/on tools Increase accuracy of automated gap filling We'll s/ll want to re- use high quality models Reproducibility / Community Models as Reusable Tools How to disseminate tens/hundreds of coupled models?

4 Microbiome Modeling in Pathway Tools Key ideas: Let's divide and conquer the problem of microbiome modeling Plug- and- play metabolic models Extensive and highly integrated so`ware support for model development, interroga/on, collabora/on, reuse We create bacterial models in 3-4 weeks Marriage of systems biology and databases Encode models as Pathway/Genome Databases (PGDBs)

5 Microbiome Modeling in Pathway Tools: Scale Speed manual development of individual models Accurate qualita/ve metabolic reconstruc/on [minutes to days] Based on MetaCyc metabolic database Enzyme name matcher tool Pathway predic/on fills many reac/on gaps Reac/on gap filler Development mode iden/fies non- produced biomass metabolites Iden/fy base blocking metabolites and the reac/ons they block Reac/on balance checker Fast interpreta/on of model results via visualiza/on tools Run models programma/cally via PythonCyc API Programming not required to build models with Pathway Tools

6 Microbiome Modeling in Pathway Tools: Collabora/ve Development / Reuse of Models Enhance model understandability Data provenance via evidence codes, comments, cita/ons, author credits Couple models with enriching informa/on Chemical structures, atom mappings, pathways, genome, regulatory network Web and desktop query and visualiza/on tools: publish models on the web Coming in 2016: run models through the web Collabora/ve model development Pathway Tools enables concurrent mul/- user upda/ng of PGDBs Transac/on history stored for PGDBs Edi/ng tools: reac/on editor, pathway editor, Marvin Share PGDBs via PGDB registry - - hip://biocyc.org/registry.html Plug- and- play models All PGDBs share reac/on and metabolite iden/fiers with MetaCyc All PGDBs share common iden/fiers for cellular compartments

7 Pathway Tools Enables Mul/- Use Metabolic Databases Encyclopedia Metabolic Model Queryable Database Zoomable Metabolic Map Omics Data Analysis

8 Pathway Tools So`ware: Workflow Annotated Genome + PathoLogic MetaFlux Pathway/Genome Database Pathway/Genome Navigator Pathway/Genome Editors Briefings in Bioinformatics 11: ,000 lines of Lisp code ~= 1.5M lines of C or Java code

9 Pathway Tools So`ware: PathoLogic Computa/onal crea/on of new Pathway/Genome Databases Transforms genome into Pathway Tools schema and layers inferred informa/on above the genome Predicts metabolic network (qualita/ve metabolic reconstruc/on) Reactome, metabolites, metabolic pathways imported from MetaCyc Batch mode for high throughput Predicts which genes code for missing enzymes in metabolic pathways Infers transport reac/ons from transporter names in genome

10 Metabolic Model Genera/on in Pathway Tools PathoLogic: Qualita/ve metabolic model reconstruc/on Inference of the reactome from the annotated genome Inference of metabolic pathways by selec/ng from MetaCyc pathways Karp et al, Stand Genomic Sci, :424 MetaFlux Infer biomass reac/on Gap fill Edit reac/on complement Iterate Karp et al, Briefings in Bioinforma4cs 2015, in press

11 Pathway Predic/on Pathway predic/on is useful because Pathways organize the metabolic network into mentally tractable units Pathways can be used for analysis of computed fluxes and high- throughput data Visualiza/on, enrichment analysis Pathway inference fills gaps in metabolic network Reduces computa/onal demands of gap filling

12 Reactome Inference For each protein in the organism, infer reac/on(s) it catalyzes Build from exis/ng genome annota/on! Match protein func/ons to MetaCyc reac/ons Enzyme names (uncontrolled vocabulary) EC numbers Gene Ontology terms

13 PathoLogic Enzyme Name Matcher Name matcher generates alterna5ve variants of each name and matches each to MetaCyc Strips extraneous informa/on found in enzyme names Puta/ve carbamate kinase, alpha subunit Flavin subunit of carbamate kinase Cytoplasmic carbamate kinase Carbamate kinase (abcd) Carbamate kinase ( )

14 MetaCyc: Curated Metabolic Database MetaCyc v Citations 45,000 Pathways 2,310 KEGG Modules SEED Subsystems Reactions 12,400 8,692 Metabolites 12,000 Minireviews: Textbook Pages 6,300 MetaCyc is free and open Contains large number of computed atom mappings A Systematic Comparison of the MetaCyc and KEGG Pathway Databases BMC Bioinformatics (1):112

15 Algorithm for Inference of Metabolic Pathways For each pathway in MetaCyc consider What frac/on of its reac/ons are present in the organism's reactome? Are enzymes present for reac/ons unique to the pathway? Is a given pathway outside its designated taxonomic range? Calvin cycle: green plants, green algae, etc Are enzymes present for designated key reac/ons within MetaCyc pathways? Calvin cycle / ribulose bisphosphate carboxylase Standards in Genomic Sciences 5:

16 Pathway Evidence Report

17 Pathway Tools So`ware: Pathway/Genome Editors Interac/vely update PGDBs with graphical editors Support geographically distributed teams of curators with object database system Gene and protein editor Reac/on editor Compound editor Pathway editor Operon editor Publica/on editor

18 MetaFlux Modeling Tool: Modes of Opera/on Solving mode Individual organisms, organism communi/es Steady- state FBA, dynamic FBA * Single compartment, 2- D spa/al grid with diffusion * Cellular- compartment aware Removal of flux loops, inference of biomass reac/on Knock- out mode (single/double gene/reac/on knock- outs) Model development mode Development mode (mul/ple gap filling) Fast Development mode (reac/on gap filling) [Latendresse 2014] Iden/fy dead- end metabolites and blocked reac/ons * Karp et al, Briefings in Bioinformatics, in press *Will be released in November 2015

19 Solver Used by MetaFlux MetaFlux formulates LP and MILP problems for SCIP solver and processes SCIP output MILP is fast Mul/- core version under development Free to academics, distributed as part of Pathway Tools Konrad- Zuse- Zentrum für Informa/onstechnik Berlin scip.zib.de

20 MetaFlux Mul/ple Gap Filling of FBA Models Reac/on gap filling (Kumar et al, BMC Bioinf :212): Reverse direc/onality of selected reac/ons AND Add minimal set of reac/ons from MetaCyc to model to enable a solu/on Reac/on cost is a func/on of reac/on taxonomic range Compartment- based gap filling available FastGap gap filler performs reac/on gap filling only, but is much faster Par/al solu/ons: Iden/fy maximal subset of biomass components for which model can yield posi/ve fluxes Metabolite gap filling: Postulate addi/onal nutrients and secre/ons

21 Other Aspects of MetaFlux Store and update metabolic model within PGDB Curate metabolic model and organism database simultaneously All query and visualiza/on tools applicable to FBA model Graphical explora/on of model results Visualize computed reac/on fluxes using cellular overview Plot organism composi/on, metabolite concentra/ons Programma/c execu/on of models from Lisp and PythonCyc APIs Other APIs exist but don't currently support modeling Perl, Java, R

22 E. coli Metabolic Model Generated from EcoCyc, updated on each EcoCyc release Predicts phenotypes of E. coli knock- outs 95.2% accuracy for 1445 genes Predicts growth/no- growth of E. coli on different nutrients 80.7% accuracy for 431 chemically defined growth media BMC Syst Biol Jun 30;8:79

23 Pain/ng E. coli Fluxes on Metabolic Map

24 Will be released in November 2015

25 E. coli Fluxes on Pathway Diagram

26 Dynamic FBA Modeling of E. coli Dynamic FBA modeling of E. coli growth under varying nutrient conditions t=1-20: E. coli grows anaerobically on 10 mmol glucose t=21-34: O 2 is added to the simulation; E. coli grows completely aerobically t=34-35: O 2 availability becomes limiting; acetate forms t=36-44: O 2 is exhausted; anaerobic growth resumes t=45 onwards: glucose is exhausted, cells begin to die

27 Dynamic Grid Modeling of a Simple Microbial Community Inspired by Segre's COMETS time io Initially, E. rectale is present throughout the grid; E. coli is present in southwest corner Halfway through simulation, B. thetaiotamicron is added to the middle of the lawn E. rectale shows higher growth where E. coli or B. theta are present because of availability of acetate from E. coli. E. rectale produces butyrate where acetate is present.

28 Curated Models in Hand Escherichia coli Eubacterium rectale Bacteriodes thetaiotamicron Methanobrevibacter smithii (in progress) Will be present in BioCyc.org and PGDB registry in November Contact me to coordinate on crea/on of addi/onal models Biocyc.org/microbiome- modeling.shtml Aiend metabolic modeling tutorials at SRI (see BioCyc.org)

29 PythonCyc: The Python API for Pathway Tools Pathway Tools API calls in Python Get PythonCyc at: latendre/pythoncyc Pathway Tools and MetaFlux plots with matplotlib

30 MetaFlux JSON output: cobrapy meets MetaFlux Load, manipulate MetaFlux models in cobrapy Snippet from EcoCyc-19.5-GEM JSON Works with any MetaFlux model (E. rectale here)

31 Acknowledgements Mario Latendresse MetaFlux implementa/on PythonCyc API Dan Weaver Model building PythonCyc API Cobrapy interface MetaCyc Collaborators Sue Rhee, Peifen Zhang Lukas Mueller, Hartmut Foerster Funding sources: NIH Na5onal Ins5tute of General Medical Sciences BioCyc webinars: biocyc.org/webinar.shtml