Pathways Databases: A Case Study in Computational Symbolic Theories

1
Pathways Databases A Case Study in Computational
Symbolic Theories

• Author Peter D. Karp

Presented by Gunjan Gupta April 13, 2004
2
Overview

• What is Gene Co-expression and why discover gene
  co-expression ?
• Using DNA microarray data for discovering
  functional gene co-expression.
• Discussion on how to use phylogenetic information
  to filter out false positives in the micro-array
  data.
• Detailed discussion of the proposed solution.

3
A Complete Scientific Theory of a simple
organism is very Complex

• A simple organism such as E. Coli is very
  complex

Example, for E. Coli - 791 chemical compounds
744 enzyme-catalyzed biochemical reactions for
the Metabolic pathways. - 4290 protein coding
genes - A perhaps larger number of mostly unknown
RNA-encoding genes
Explains all the observed processes in an
organism
4
Too complex(large) for a single Scientist to grasp

• More complex than a factory producing cars

Imagine a car factory that not only produces
cars, but produces other factories of its own
kind, and everything else needed to produce,
reproduce, repair, fend off competition, fight
foreign attacks, adapt to a changing environment,
evolve !
5
Traditional methods of representing theoretical
knowledge

• Technical papers in journals
• Diagrams, flowcharts tables
• Mostly natural language descriptions more
  suitable for humans.
• Raw and partially processed data in large
  biological databases example, indexed by search
  criteria, annotated in English etc.

6
Traditional methods not enough, why ?

• Cannot represent scientific theories for
  biological systems too complex.
• Automatic deductions not possible as the theory
  is not stored in a structured, computer readable
  presentation.
• Natural Language Processing not developed to the
  level yet to allow a computer to search human
  readable data.
• Only bits and pieces understandable by one person
  at any given time.
• Large-scale patterns and deductions might never
  be found in the absence of any one individual who
  can see the big picture.
• Addition of details from experimentation does not
  necessarily improve our understanding of the
  whole system.

7
Pathway Databases for Biology

• A Database of Scientific theory storing
  qualitative information semantics of the
  theory in a well-defined form.
• Useful for representing theories that are mostly
  non-numerical in nature a biological system is
  well suited.
• Definition A Pathway is a linked set of
  biochemical reactions, linked as follows

Theory of relationships between 1,2,3,4,5,6 7
defined)
Reaction 6
Reaction 1
Reaction 2
Chemical 8
Raw Material 5
8
Different Kinds of Biological Pathways Database
(PDB)

• Only Metabolic Pathways (majority) (time
  generally fast)
• e.g. Production of ATP in the Mitochondria
• Signaling Pathways (time generally medium)
• Intra-cellular e.g. cell-membrane to nucleus
• Extra-cellular e.g. growth-factors for nerve,
  skin, say for injury..
• Genetic-regulatory pathways (time generally
  medium/slow)
• gene expression control, make more as needed,
  example synthesize amino-acids deficient in food
  by increasing expression of related genes.
• A combination of one or more of the above (e.g.
  PAD, EcoCyc).

9
Pathways Databases an intersection of 4 fields
Genomics
Biochemistry
Databases
Artificial Intelligence
10
Examples of current PDBs

• Metabolic Pathway Database
• EcoCyc (author) http//www.ecocyc.org/
• Signaling Pathways
• SPAD http//www.grt.kyushu-u.ac.jp/spad/
• Genetic-regulatory pathways
• BIOBASE http//www.gene-regulation.com/

11
EcoCyc Project

• EcoCyc Encyclopedia of Escherichia coli K12
  Genes and Metabolism
• At http//www.ecocyc.org/
• Links to a bigger site http//biocyc.org/server.
  html containing PDB for other organisms including
  human.

12
What Metabolic Genetic Pathway Info does EcoCyc
Contain ?

• For each Enzyme in the PDB
• Detailed description of reaction catalyzed by
  each enzyme.
• Genes to which the enzymes map to, if available.
• The range of substrate the enzyme would accept.
• Chemicals that inhibit or activate the Enzyme.
• Its subunit structure.
• Each small molecule enzyme substrate
• Pathways types for E-Coli included
• Biosynthesis of cellular bulding blocks.
• Extraction of Carbon from food.
• Extraction of chemical energy from food.

13
Tools Visualization in EcoCyc

• SRI International (http//www.sri.com/) developed
  the visualization and search tool called Pathway
  Tools as an intelligent user interface.
• Allows the user to exploits the semantic
  information in the PDB and write complex queries.
• Visualize results as a Pathway graph called
  Overview Diagram.
• A variety of criteria for the queries
• Name matching
• Classification hierarchy (taxonomy, metabolic
  pathways)
• Example
• Find all reactions that are activated or
  inhibited by a given metabolite.
• Superimposing genetics data on the visualization
  (see demo)
• User can create a PGDB for a new organism and
  share it.

14
Visualization
Metabolite categories
Individual Reactions
Glycolysis region
15
An application automated inter-species
comparison of reactions

• Yellow shows reactions in E. Coli that match with
  another species S. Cerevisiae, found as a
  result of database lookup.
• Mostly automated layout using Pathway Tools (some
  manual fitting was done by author in this paper
  for this diagram).

16
Exploiting Knowledge Representation AI tools in a
PDB

• Building an Ontology DB schema defining the
  precise relationships between entity use UML ?
  ?
• Encode a theory using the Ontology
• EcoCyc ontology consists of 1000 object-oriented
  classes encoding key concepts of biology and
  biochemistry.
• Extend Ontology when new concepts are found that
  cannot be derived using existing Ontology.
• Use KR techniques that exploit Symbolic AI
  reasoning to say
• build an inference engine (see Bruce Porters
  Knowledge Machine for example) on the PDB or
• Perform specific global inferences using specific
  relationship searches.

17
Example of Symbolic Inferences on EcoCyc

• Results of a search changed the simplistic notion
  of what gene is
• In E. Coli found 1 out of 7 enzymes catalyze more
  than one reaction, and almost 1 out of 7 cases
  where an reaction is used in more than one
  Pathway (overlapping sub-graphs or clusters)
  this can be treated as a discovered theory.
• Characterized the transcription factors
  relationship (see next slide).
• Other interesting theories found using PDB
• Scale free network topology that follows a power
  law for both Metabolic and Genetic networks.
• Deletion of proteins with high connectivity more
  likely to kill the organism.

18
Example 2 Characterizing Transcription Factors
inter-relationship in a genetic pathway
Just two dominate the relationships making the
tree very shallow
Most do not regulate themselves or other
transcription factors
19
EcoCyc Demo

• Demo 1 Search demo
• Go to site http//www.ecocyc.org/
• Click on DB search
• Search for Glycolysis and show to class if time
  permits ..
• Demo 2 Combining pathways with Gene expression
  data
• Go to site http//www.ecocyc.org1555/expression.h
  tml
• Specify data file as http//biocyc.org/coli.dat
  (might have to save it locally to work).
• Select absolute display level.
• Enter ratio for numerator as 1.

20
Issues/Comments

• The idea itself is quite powerful combining
  sequence data with DNA array, and results seem to
  be quite good, but ..
• Too many steps hard to quantify the results
  except empirically, because of the complexity. At
  least 5 levels of successive transformations
  (1.genes to meta-genes via blast 2. Pearson
  correlation 3. order based probability 4. network
  thresholding 5. Clustering).
• Some heuristics are not explained much for
  clustering, transformation into 2-d space from
  probability for example, where the original
  problem was a graph- why not directly partition
  the graph ?
• Limited quality of clusters because of 2-d
  translation. Not clear why and how the data fits
  into a 2-d space. Not clear if the translation
  using P-value is a metric.
• The paper was obviously not written by plain
  computer scientists - lot of interesting
  discoveries and analysis after the method was
  used.

21
One line summary of the paper ..
Using phylogenetics to filter out gene
co-expressions in micro-array data that are not
functionally relevant.