PathoLogic Pathway Predictor

1
PathoLogic Pathway Predictor
2
Inference of Metabolic Pathways
Annotated Genomic Sequence
Pathway/Genome Database
Pathways
Reactions
PathoLogic Software Integrates genome and pathway
data to identify putative metabolic networks
Compounds
Multi-organism Pathway Database (MetaCyc)
Gene Products
Genes
Genomic Map
3
PathoLogic Functionality

• Initialize schema for new PGDB
• Transform existing genome to PGDB form
• Infer metabolic pathways and store in PGDB
• Infer operons and store in PGDB
• Assemble Overview diagram
• Assist user with manual tasks
• Assign enzymes to reactions they catalyze
• Identify false-positive pathway predictions
• Build protein complexes from monomers
• Infer transport reactions

4
PathoLogic Input/Output

• Inputs
• File listing genetic elements
• http//bioinformatics.ai.sri.com/ptools/genetic-el
  ements.dat
• Files containing DNA sequence for each genetic
  element
• Files containing annotation for each genetic
  element
• MetaCyc database
• Output
• Pathway/genome database for the subject organism
• Reports that summarize
• Evidence contained in the input genome for the
  presence of reference pathways
• Reactions missing from inferred pathways

5
PathoLogic Analysis Phases

• Trial parsing of input data files few days
• Initialize schema of new PGDB 3 min
• Create DB objects for replicons, genes, proteins
  5 min
• Assign enzymes to reactions they catalyze
• ferrochelatase
  10 min / 1 week
• glutamate 1-semialdehyde 2,1-aminomutase
• porphobilinogen deaminase

E1
E2
B
D
E
F
6
PathoLogic Analysis Phases

• From assigned reactions, infer what pathways are
  present
  5 min / few days
• Define metabolic overview diagram 30
  min
• Define protein complexes
  few days

7
genetic-elements.dat

• ID TEST-CHROM-1
• NAME Chromosome 1
• TYPE CHRSM
• CIRCULAR? N
• ANNOT-FILE chrom1.pf
• SEQ-FILE chrom1.fsa
• //
• ID TEST-CHROM-2
• NAME Chromosome 2
• CIRCULAR? N
• ANNOT-FILE /mydata/chrom2.gbk
• SEQ-FILE /mydata/chrom2.fna
• //

8
File Naming Conventions

• One pair of sequence and annotation files for
  each genetic element
• Sequence files FASTA format
• suffix fsa or fna
• Annotation file
• Genbank format suffix .gbk
• PathoLogic format suffix .pf

9
Typical Problems Using Genbank Files With
PathoLogic

• Wrong qualifier names used read PathoLogic
  documentation!
• Extraneous information in a given qualifier
• Check results of trial parse carefully

10
GenBank File Format

• Accepted feature types
• CDS, tRNA, rRNA, misc_RNA
• Accepted qualifiers
• /locus_tag Unique ID recm
• /gene Gene name req
• /product req
• /EC_number recm
• /product_comment opt
• /gene_comment opt
• /alt_name Synonyms opt
• /pseudo Gene is a pseudogene opt
• For multifunctional proteins, put each function
  in a separate /product line

11
PathoLogic File Format

• Each record starts with line containing an ID
  attribute
• Tab delimited
• Each record ends with a line containing //
• One attribute-value pair is allowed per line
• Use multiple FUNCTION lines for multifunctional
  proteins
• Lines starting with are comment lines
• Valid attributes are
• ID, NAME, SYNONYM
• STARTBASE, ENDBASE, GENE-COMMENT
• FUNCTION, PRODUCT-TYPE, EC, FUNCTION-COMMENT
• DBLINK
• INTRON

12
PathoLogic File Format

• ID TP0734
• NAME deoD
• STARTBASE 799084
• ENDBASE 799785
• FUNCTION purine nucleoside phosphorylase
• DBLINK PIDg3323039
• PRODUCT-TYPE P
• GENE-COMMENT similar to GP1638807 percent
  identity 57.51 identified by sequence
  similarity putative
• //
• ID TP0735
• NAME gltA
• STARTBASE 799867
• ENDBASE 801423
• FUNCTION glutamate synthase
• DBLINK PIDg3323040
• PRODUCT-TYPE P

13
Before you start What to do when an error occurs

• Most Navigator errors are automatically trapped
  debugging information is saved to error.tmp file.
• All other errors (including most PathoLogic
  errors) will cause software to drop into the Lisp
  debugger
• Unix error message will show up in the original
  terminal window from which you started Pathway
  Tools.
• Windows Error message will show up in the Lisp
  console. The Lisp console usually starts out
  iconified its icon is a blue bust of Franz
  Liszt
• 2 goals when an error occurs
• Try to continue working
• Obtain enough information for a bug report to
  send to pathway-tools support team.

14
The Lisp Debugger

• Sample error (details and number of restart
  actions differ for each case)
• Error Received signal number 2 (Keyboard
  interrupt)
• Restart actions (select using continue)
• 0 continue computation
• 1 Return to command level
• 2 Pathway Tools version 10.0 top level
• 3 Exit Pathway Tools version 10.0
• 1c EC(2)
• To generate debugging information (stack
  backtrace)
• zoom count all
• To continue from error, find a restart that takes
  you to the top level in this case, number 2
• cont 2
• To exit Pathway Tools
• exit

15
How to report an error

• Determine if problem is reproducible, and how to
  reproduce it (make sure you have all the latest
  patches installed)
• Send email to ptools-support_at_ai.sri.com
  containing
• Pathway Tools version number and platform
• Description of exactly what you were doing (which
  command you invoked, what you typed, etc.) or
  instructions for how to reproduce the problem
• error.tmp file, if one was generated
• If software breaks into the lisp debugger, the
  complete error message and stack backtrace
  (obtained using the command zoom count all, as
  described on previous slide)

16
Using the PPP GUI to Create a Pathway/Genome
Database

• Input Project Information
• Organism -gt Create New

17
Input Project Information
18
PathoLogic Command Menus

• Organism
• Select
• Create New
• Save KB
• Revert KB
• Reinitialize KB
• Specify Reference PGDB(s)
• Exit
• Build
• Trial Parse
• Automated Build

• Refine
• Assign Probable Enzymes
• Assign Modified Proteins
• Create Protein Complexes
• Re-run Name Matcher
• Rescore Pathways
• Predict transcription units
• Transport Identification Parser
• Update Overview
• Pathway Hole Filler

19
Next Steps

• Trial Parse
• Build -gt Trial Parse
• Fix any errors in input files
• Build pathway/genome database
• Build -gt Automated Build

20
PathoLogic Parser Output
21
Assign Enzymes to Reactions
5.1.3.2
Gene product
MetaCyc
UDP-glucose-4-epimerase
Match
yes
no
Assign
Probable enzyme -ase
UDP-D-glucose ? UDP-galactose
no
yes
Manually search
Not a metabolic enzyme
yes
no
Assign
Cant Assign
22
Enzyme Name Matcher

• Matches on full enzyme name
• Match is case-insensitive and removes the
  punctuation characters -_()',
• Also matches after removal of prefixes and
  suffixes such as
• Putative, Hypothetical, etc
• alphabetacatalyticinducible
  chainsubunitcomponent
• Parenthetical gene name

23
Enzyme Name Matcher

• For names that do not match, software identifies
  probable metabolic enzymes as those
• Containing ase
• Not containing keywords such as
• sensor kinase
• topoisomerase
• protein kinase
• peptidase
• Etc
• Research unknown enzymes
• MetaCyc, Swiss-Prot, PubMed

24
Enzyme Name to Reaction Mapping
See also file PTools Tutorial/PathoLogic
Reports/name-matching-report.txt
25
Manual Polishing

• Refine -gt Assign Probable Enzymes ? Do this
  first
• Refine -gt Rescore Pathways ?
  Redo after assigning enzymes
• Refine -gt Create Protein Complexes ? Can be
  done at any time
• Refine -gt Assign Modified Proteins ? Can
  be done at any time
• Refine -gt Transport Identification Parser ? Can
  be done at any time
• Refine -gt Pathway Hole Filler
• Refine -gt Predict Transcription Units
• Refine -gt Update Overview ? Do this last, and
  repeat after any material changes to PGDB

26
Assign Probable Enzymes
27
How to find reactions for probable enzymes

• First, verify that enzyme name describes a
  specific, metabolic function
• Search for fragment of name in MetaCyc you may
  be able to find a match that PathoLogic missed
• Look up protein in SwissProt or other DBs
• Search for gene name in PGDB for related organism
  (bear in mind that gene names are not reliable
  indicators of function, so check carefully)
• Search for function name in PubMed
• Other

28
Manual Polishing

• Refine -gt Assign Probable Enzymes
• Refine -gt Rescore Pathways
• Refine -gt Create Protein Complexes
• Refine -gt Assign Modified Proteins
• Refine -gt Transport Identification Parser
• Refine -gt Pathway Hole Filler
• Refine -gt Predict Transcription Units
• Refine -gt Run Consistency Checker
• Refine -gt Update Overview

29
Automated Pathway Inference

• All pathways in MetaCyc for which there is at
  least one enzyme identified in the target
  organism are considered for possible inclusion.
• Algorithm errs on side of inclusivity easier to
  manually delete a pathway from an organism than
  to find a pathway that should have been predicted
  but wasnt.

30
Considerations taken into account when deciding
whether or not a pathway should be inferred

• Is there a unique enzyme an enzyme not involved
  in any other pathway?
• Does the organism fall in the expected taxonomic
  domain of the pathway?
• Is this pathway part of a variant set, and, if
  so, is there more evidence for some other
  variant?
• If there is no unique enzyme
• Is there evidence for more than one enzyme?
• If a biosynthetic pathway, is there evidence for
  final reaction(s)?
• If a degradation pathway, is there evidence for
  initial reaction(s)?
• If an energy metabolism pathway, is there
  evidence for more than half the reactions?

31
Assigning Evidence Scores to Predicted Pathways

• XYZ denotes score for P in O
• where
• X total number of reactions in P
• Y enzymes catalyzing number of reactions for
  which there is evidence in O
• Z number of Y reactions that are used in other
  pathways in O

32
Manual Pruning of Pathways

• Use pathway evidence report
• Coloring scheme aids in assessing pathway
  evidence
• Phase I Prune extra variant pathways
• Rescore pathways, re-generate pathway evidence
  report
• Phase II Prune pathways unlikely to be present
• No/few unique enzymes
• Most pathway steps present because they are used
  in another pathway
• Pathway very unlikely to be present in this
  organism
• Nonspecific enzyme name assigned to a pathway
  step

33
Caveats

• Cannot predict pathways not present in MetaCyc
• Evidence for short pathways is hard to interpret
• Since many reactions occur in multiple pathways,
  some false positives

34
Output from PPP

• Pathway/genome database
• Summary pages
• Pathway evidence page
• Click Summary of Organisms, then click organism
  name, then click Pathway Evidence, then click
  Save Pathway Report
• Missing enzymes report
• Directory tree containing sequence files,
  reports, etc.

35
Resulting Directory Structure

• ROOT/ptools-local/pgdbs/user/ORGIDcyc/VERSION/
• input
• organism.dat
• organism-init.dat
• genetic-elements.dat
• annotation files
• sequence files
• reports
• name-matching-report.txt
• trial-parse-report.txt
• kb
• ORGIDbase.ocelot
• data
• overview.graph
• released -gt VERSION

36
Manual Polishing

• Refine -gt Assign Probable Enzymes
• Refine -gt Rescore Pathways
• Refine -gt Create Protein Complexes
• Refine -gt Assign Modified Proteins
• Refine -gt Transport Identification Parser
• Refine -gt Pathway Hole Filler
• Refine -gt Predict Transcription Units
• Refine -gt Run Consistency Checker
• Refine -gt Update Overview

37
Creating Protein Complexes
38
Complex Subunits Stoichiometries
39
Manual Polishing

• Refine -gt Assign Probable Enzymes
• Refine -gt Re-run Name Matcher
• Refine -gt Create Protein Complexes
• Refine -gt Assign Modified Proteins
• Refine -gt Transport Identification Parser
• Refine -gt Pathway Hole Filler
• Refine -gt Predict Transcription Units
• Refine -gt Run Consistency Checker
• Refine -gt Update Overview

40
Proteins as Reaction Substrates
41
Manual polishing

• Refine -gt Assign Probable Enzymes
• Refine -gt Re-run Name Matcher
• Refine -gt Create Protein Complexes
• Refine -gt Assign Modified Proteins
• Refine -gt Transport Identification Parser
• Refine -gt Pathway Hole Filler
• Refine -gt Predict Transcription Units
• Refine -gt Run Consistency Checker
• Refine -gt Update Overview

42
What are pathway holes?
At least one reaction in the pathway has an
enzyme assigned. The reactions in the pathway
without enzymes assigned are holes.
1.4.3.-
iminoaspartate
No EC
L-aspartate
quinolinate
holes
n.n. pyrophosphorylase nadC, RV1596
6.3.1.5
deamido-NAD
deamido-NAD
nicotinate nucleotide
2.7.7.18
6.3.5.1
NAD
43
Algorithm for identifying candidates and
consolidating data
Step III IV Consolidate hits and evaluate
evidence using a Bayes classifier
Step II BLAST against target genome
Step I collect query isozymes of function A
3 queries have low-scoring hits to sequence
X Resulting P(has-function) is low
gene X
organism 1 enzyme A
organism 2 enzyme A
organism 3 enzyme A
8 queries have high-scoring hits to sequence
Y Resulting P(has-function) is high
organism 4 enzyme A
organism 5 enzyme A
gene Y
organism 6 enzyme A
organism 7 enzyme A
organism 8 enzyme A
5 queries have low-scoring hits to sequence
Z Resulting P(has-function) is low
gene Z
target genome
44
Reference for the Pathway Hole Filler

• Green, ML and Karp, PD. A Bayesian method for
  identifying missing enzymes in predicted
  metabolic pathway databases. BMC Bioinformatics
  2004, 576.

45
Features used to calculate the probability that a
protein has the desired function
Candidate is in a contiguous set of genes
transcribed in one direction with another gene in
the pathway

• Best E-value
• Avg. rank
• Avg aligned
• Number of query sequences aligned
• Potential operon?
• Adjacent reactions?

Candidate is adjacent to the gene assigned to an
adjacent reaction in the pathway
46
Navigating to the Pathway Hole Filler
47
Steps that must be completed before running the
Pathway Hole Filler

• Install BLAST executable (should already be
  installed on training room machines)
• Prepare BLAST protein db
• Need FASTA format genome nucleotide sequence (see
  instructor if you have something different, like
  ESTs, or have no sequence data file)
• In general, the more pathways in your PGDB, the
  more the pathway hole filler will have to search
  for

48

• Steps for operating the pathway hole filler
• Prepare training data for Bayes classifier
• Collect feature data for known rxns in PGDB
• Calculate probability distributions for classifier

• Identify and evaluate candidates
• Collect feature data for each candidate
• Use classifier to determine P(has-function)

• Choose holes to fill in KB
• Either select all above a cutoff or manually
  review candidates

49
Step 1 Prepare Training Data

• Calculate training data from your organism or use
  existing training data

• Once Step 1 has been completed, the training data
  are saved and can be reused (even in another
  Pathway Tools session).
• If using existing data from E. coli the training
  data are based on data from the literature.

50
Step 2 Identify Evaluate Candidates
51
Step 2 Identify Evaluate Candidates
A list of all pathway holes in the PGDB
A list of all pathways in the PGDB with holes
52
Modes of operation

• Fully automatic
• No interaction required from user
• All default values used
• Prepare training data all known rxns in KB
• Identify and evaluate candidates all pathways
  with pathway holes
• Choose holes to fill in KB all holes with Pgt0.9
  filled

53
Modes of operation

• Wizard
• Wizard prompts user for training data source and
  for which holes to make predictions. Wizard runs
  Steps 1 2, then prompts user to complete Step 3.

Power-user mode User must proceed through each
step in order. Program still prompts user for
required parameters, but each step must be
completed before advancing to next step.
54
Step 3 Choose Holes to Fill in KB
55
Step 3 Choose Holes to Fill in KB
56
(No Transcript)
57
(No Transcript)
58
(No Transcript)
59
Output from Pathway Hole Filler- from Prepare
Training Data step

• ROOT/aic-export/ecocyc/ORGIDcyc/VERSION/data/
• (e.g., ROOT/aic-export/ecocyc/caulocyc/1.0/data/)
• rxn-list data retrieved from ORGID for
  calculating training data
• priors/ directory containing training data that
  is loaded when using existing data from ORGID
• These files contain the training data computed in
  Step 1. If either file is available, the user may
  use existing training data in Step 1.
• Each file is overwritten each time you run this
  step.

60
Output from Pathway Hole Filler- from Identify
and Evaluate Candidates step

• ROOT/aic-export/ecocyc/ORGIDcyc/VERSION/reports/
• (e.g., ROOT/aic-export/ecocyc/caulocyc/1.0/reports
  /)
• ORGIDholesX-Y.html (e.g., CAULOholes0-10.html)
• ORGID_filled-holes.html the list of holes that
  user selected to fill in the KB in Step 3.
• blasterrors.log log of each rxn describing
  whether or not any candidates were found
• hole-data file containing data (in a Lisp
  structure) found for each rxn, used to generate
  list in Choose holes to fill in KB dialogue. If
  this file is available, step 3 can be initiated
  without repeating Step 2.
• Each file is overwritten each time you run this
  step.

61
Manual polishing

• Refine -gt Assign Probable Enzymes
• Refine -gt Rescore Pathways
• Refine -gt Create Protein Complexes
• Refine -gt Assign Modified Proteins
• Refine -gt Transport Identification Parser
• Refine -gt Pathway Hole Filler
• Refine -gt Predict Transcription Units
• Refine -gt Run Consistency Checker
• Refine -gt Update Overview

62
Nomenclature

• WO pair pair of genes within an operon
• TUB pair pair of genes at a transcription unit
  boundary (delineate operons)

63
Operation of the operon predictor

• For each contiguous gene pair, predict whether
  gene pairs are within the same operon or at a
  transcription unit boundary
• Use pairwise predictions to identify potential
  operons
• AB TUB pair
• BC WO pair operon BCD
• CD WO pair
• DE TUB pair

A
B
C
D
E
64
Operon predictor

• Predicts operon gene pairs based on
• intergenic distance between genes
• genes in the same functional class
• Typically used for operon prediction
• We use method from Salgado et al, PNAS (2000) as
  a starting point.
• Uses E. coli experimentally verified data as a
  training set.
• Compute log likelihood of two genes being WO or
  TUB pair based on intergenic distance.

65
Operon predictor

• Additional features easily computed from a PGDB
• both genes products enzymes in the same metabolic
  pathway
• both gene products monomers in the same protein
  complex
• one gene product transports a substrate for a
  metabolic pathway in which the other gene product
  is involved as an enzyme
• a gene upstream or downstream from the gene pair
  (and within the same directon) is related to
  either one of the genes in the pair as per
  features 1, 2 and 3 above.