Meaningful Metabolic Logic Lynda B.M. Ellis Laboratory Medicine and Pathology University of Minnesota People Top: Larry Wackett, Michael Turnbull, Kathrin Fenner, Ted LaBelle Bottom: Junfeng (Jeff) Gao, Chunhui Li, Rachael Long, Lynda Ellis The University of Minnesota Biocatalysis/Biodegradation Database (UM-BBD) Freely available on the World Wide Web http://umbbd.msi.umn.edu/ Focus: Microbial specialized catabolism Emphasizing environmental pollutants Represents metabolism as pathways From 1,2-Dichloroethane to 2-Chloroethanol Graphic of the reaction. Medline reference Verschueren KH, Seljee F, Rozeboom HJ, Kalk KH, Dijkstra BW Nature (1993) 363(6431): 693-8. Search Medline titles for haloalkane dehalogenase. 64 citations found on August 14, 2005. 1,2-Dichloroethane | | H2O haloalkane | / dehalogenase |/ 184.108.40.206 | Search GenBank, 89 hits on Aug. 14, 2005 Kyoto |\ ExPASy | \ | HCl v 2-Chloroethanol Reaction Page Generate a pathway starting from this reaction. UM-BBD Biotransformation rules in accord with this reaction: Alkyl halide -----> Alcohol (bt0022) ______________________________________________________________________ [1,2-Dichloroethane] [BBD Main Menu] Page Author(s): Renhao Li August 14, 2005 Contact Us This is the UM-BBD reaction, reacID# r0001. It was generated on August 24, 2005 2:42:28 PM CDT. © 2005, University of Minnesota. Metapathway Map http://umbbd.msi.umn.edu/ meta/meta_map.html Hypothesis ~107 compounds in the environment. The UM-BBD will never contain all such metabolism. UM-BBD data might be used to predict reasonable biodegradation schemes for compounds it does not contain. Prediction Overview http://umbbd.msi.umn.edu/predict/ Biotransformation Rule vic-dihydroxybenzenoid -> extradiol ring cleavage Aerobic Likelihood: Likely UM-BBD Reaction(s): 2'-Aminobiphenyl-2,3-diol -----> 2-Hydroxy-6-oxo-(2'-aminophenyl)hexa-2E,4Z-dienoate (reacID# r0457) Catechol -----> 2-Hydroxy-cis,cis-muconate semialdehyde (reacID# r0307) 3-Chlorocatechol -----> 3-Chloro-2-hydroxymuconic semialdehyde (reacID# r1022) 4-Chlorocatechol -----> 5-Chloro-2-hydroxymuconic semialdehyde (reacID# r0287) … Contact Us if you have any comments on rule bt0008. Biotransformation Rule-base bt0001: primary alcohol aldehyde bt0002: secondary alcohol ketone bt0003: aldehyde carboxylate bt0005: vic-unsubstituted aromatic cis-dihydroxydihydroaromatic bt0008: vic-dihydroxybenzenoid extradiol ring cleavage … bt0255: vic-dihydrodihydroxyaromatic vic-dihydroxyaromatic bt0259: vic-dihydroxyaromatic intradiol ring cleavage Pathway Prediction Example Start After five steps Complexity - 100gm of soil • 100,000,000,000 bacteria • 100,000 natural chemicals • 10,000 bacterial species • 1 new chemical PredictBT Workshops Since 1998, three workshops held Fourth, May 4-5, 2007 Second workshop (May, 2005) assigned aerobic likelihood to btrules http://umbbd.msi.umn.edu/predictbt/ Aerobic Likelihood Standard conditions Aerobic Soil (moderate moisture) or water 25 C No other toxic or competing chemicals Rank Rules: Very Likely, Likely, Neutral, Unlikely, Very Unlikely, Unknown Expert Rankings 7 experts; 6 assigned 50 of the 250 rules, one ranked all 250. Each rule ranked by 2-3 experts prior to the workshop (prework), on a 1-5 scale Unknown rarely used (1-2 per expert) Most rankings within 1 unit If so, used rank closest to 3 (2,3 = 3; 4,5 = 4). If no, discussed at workshop, consensus reached 2 experts agreed to rank future rules Challenges of Likelihood Ranking Are two experts enough? Only 21 of 230 rules based on 10+ reactions “Them as has, gets” - not likely to improve How to rank rules for less common reactions? Challenges, 2 Errors R-COOH + CoA CoA thioester (unlikely) divided into aliphatic (likely) and aromatic (unlikely) Peroxide alcohol (very likely) (neutral) cis-dihydroxyhydrocarboxylate formation Single ring reactions were Likely This fused ring reaction was Unlikely Now all single and fused rings are Likely Challenges, 3 Over half rules ranked “Neutral” 140 Summary of BT rule rankings 128 120 100 Number of BT rules 80 60 48 40 17 10 32 20 0 very likely likely neutral unlikely very unlikely Relative Like lihood Challenges, 4 Based on BBD reactions Oxidative cleavage of secondary amines (bt0063) Unlikely, based on BBD exotic examples Simpler ones are likely How to separate? Carbamyl degradation - known, but no BBD or KEGG examples. No btrule. Should non-BBD-based rules be written? Challenges, 5 Hydrolytic vs reductive dehalogenation Hydrolytic dehalogenation aerobically likely, unless a highly halogenated compound “Highly”? X/C ratio >=1 ? H Cl Cl Cl Cl H Cl H H H Cl H H Cl Cl H Cl Cl H H H H Cl Cl aerobic hydrolytic dehalogenation anaerobic reductive dehalogenation Challenges, 6 Decarboxylation (bt0051) Too general, how to limit? 27 example BBD reactions; more in KEGG >10 different reaction patterns Instead, excluded fatty acids Other exclusions for this rule? Exclusions for other rules? Challenges, 7 Change in software breaks rules Change in hardware breaks rules Change in personnel breaks rules Data entry Rule writers Programmers Challenges, 8 Interactive web-based application Predictions must be fast Cpd button shows exact match Smlr button could show similar compounds; but is it quick enough? All rule patterns are tested against each query; each additional pattern slightly slows predictions Combinatorial Explosion Parent compound # possible metabolites # possible pathways Glyphosate (RoundUp™) Diuron (phenylureas) Nitrilotriacetate 14 27 22 132 2,223 8,444 MTBE (methy-t-butyl ether) Atrazine 148 280 1,977,448 374,052,438 Kathrin Fenner, 2007, personal communication 2007 PredictBT Workshop May 4-5, 2007, Cargill Building, UM Goal - to limit PPS predictions in a meaningful, efficient, effective way http://umbbd.msi.umn.edu/predictbt/ Sponsors: 2007 PredictBT Panel John Bumpus Stefan Kramer U Northern Iowa Kathrin Fenner EAWAG, Switzerland Susumu Goto Kyoto U, Japan TUM, Germany Fangping Mu Los Alamos Nat Lab Mukesh Patel Lhasa Limited, UK Jeanne VanBriesen Carneige-Mellon U PredictBT Suggestions Ask experts to rerank “neutral” rules Would need to rerank all No funding for this Create list of Readily Degradable Intermediates (RDIs) Develop separate, batch, mode Create complete tree Prune Even two levels useful; more chances to find an RDI PredictBT Suggestions, 2 Prune tree using thermodynamics Based on ATP generation, known pathways are in the top 20 percentile of possible pathways. Prune tree using RDIs Find the shortest path to an RDI. PredictBT Suggestions, 3 Susumo Goto: Use similarity biological, not chemical, similarity NH2 O NH2 O ≈ >.9 Tanimoto O O Different RDIs PredictBT Suggestions, 4 Use chemical similarity to prescreen queries If no BBD compounds are chemically similar to the query, PPS predictions are less accurate Use machine learning to determine which biotransformations are most likely for a given substructure. Need positive and negative examples Need rules based on 10+ reactions 8/21 are problematic rules MEPPS Started in 2002, funded by Lhasa Limited PPS rules Meteor, forming MEPPS. Meteor Environmental Pathway Prediction System Static UM-BBD web pages MEPPS. MEPPS links to static UM-BBD btrule web pages via MEPPS Biotransformation screen “Examples” button. Aerobic likelihoods added using existing Meteor metabolic reasoning tools. MEPPS Prediction NEUTRAL MEPPS UM-BBD Future Implement PredictBT workshop suggestions Add pathways rules features Improve guidance for users Develop MEPPS in parallel with PPS Questions? Workshop suggestions & MEPPS? Priority of workshop suggestions? Other comments?