Software Tools in Systems Biology

1
Software Tools in Systems Biology
2
History of Metabolic Simulation

• Analog Computers
• The earliest recorded reaction network simulation
  is from
• a paper by Chance et al, in 1943 where they
  simulated
• the peroxidase enzyme.
• But in 1943, there werent any digital computers

3
Analog Computers

• An analog computer operates quite differently
  from a digital
• computer. For a start, all operations in an
  analog computer are
• performed in parallel. Secondly, data are
  represented in an
• analog computer as voltages, a very compact but
  not
• necessarily robust form of storage (prone to
  noise corruption).
• A single capacitor (equivalent to the digital
  computers use of a
• transistor) in an analog computer can represent
  one continuous
• variable.

4
Analog Computers

• The world's first analog computer was built in
  ..??

5
Analog Computers

• The world's first analog computer was built in
  approximately 80 BC
• Found in 1901 by sponge divers near Crete, was a
  device, called the
• Antikythera Mechanism, which it is assumed to
  have been used to to show
• the motion of the Moon, Sun, and most likely the
  Planets through the
• years.

6
Analog Computers

• The world's first electronic analog computer
  (1930s) filled a room at the
• Massachusetts Institute of Technology. Vannevar
  Bush's differential
• analyzer crunched through calculus in seconds,
  although technicians often
• spent hours setting it up to solve an equation. A
  labyrinth of wheels, discs,
• shafts and gears handled the brainwork with
  precision unmatched by any
• contemporaneous machine. But its heyday was
  short, as digital computers
• developed after World War II put the analyzer out
  to pasture.

7
Analog Computers

• EC-1 Educational Analog Computer, 1961
• The Heathkit Educational Analog Computer is
  completely self-contained
• and contains nine DC operational amplifiers with
  provision for balancing
• without removing problem setup. It also features
  three initial condition
• power supplies, five coefficient potentiometers,
  four sets of relay contacts,
• an electronically regulated power supply and a
  built-in repetitive oscillator
• for automatic operation. The complete EC-1 kit
  also contains
• an assortment of precision resistors, capacitors,
  special silicon diodes and
• patch cords for setting up scores of complex
  computer problems easily
• and accurately.

8
Analog Computers
dy/dt -k y
This equation describes the breakdown of a
substance, y, with rate constant k
9
Analog Computers
Advantages of Analog Computers Very fast,
computations are performed in parallel It is
much easier to do interactive simulations because
the analog computer reacts instantly to any
parameter changes. Note The summation theorem in
MCA was discovered on an analog computer.
10
Analog Computers
Disadvantages Prone to noise Difficult to set
up large models By the 1960s, Analog computer
could not compete with the new digital computers.
11
The Digital Age
The first reaction network computer software was
called BIOSSIM and was written by David Garfinkel
written in FORTRAN and run in batch mode. Many
of the problems associated with digital
simulation were solved during this period.
12
The Digital Age
By the end of the 70s, personal computers began
to become available and started a new phase in
biochemical reaction network simulations. SCAMP
was one of the first, if not the first of a
entire new generation of simulators that became
available from the very early 80s and into the
90s. Today there are many packages to choose
from, and more are still being written.
13
Research Based Simulators

• JDesigner Mix Delphi/C/FORTRAN
• Jarnac Mix Delphi/C/FORTRAN
• SCAMP Mix Delphi/C
• SBW/SBML Mix Delphi/C/C/Java
• Gepasi Mix C/C/FORTRAN
• VCell Java, FORTRAN (?)
• bioSPICE Mix Everything!
• E-Cell C/C, FORTRAN (?)
• jigCell Mix Java/XPP
• Cellerator Mathematica
• NetBuilder C (uses third-party lib)

14

• BIOSSIM (1968)
• ESSYN (1976)
• SCAMP (1983)
• SCOP (1986)
• METAMOD (1986)
• SIMFIT (1990)
• METAMODEL (1991)
• METASIM (1992)
• KINSIM (1993)
• GEPASI (1994)
• METALGEN (1994 ?)
• MIST (1995)
• METABOLIKA (1997 ?)
• METAFLUX (1997)
• SIMFLUX (1997)
• MNA (1998)
• CELLMOD (1998)
• FLUXMAP (1999)
• METATOOL (1999)

Modelling Tools
Period
Klaus Mauch, University of Stuttgart
15
Commercial Software

• Physiome (Dead)
• Entelos
• Gene Network Sciences
• Genomatica
• TeraNode

16
Commercial Software

• 3rd Millennium, Accelrys, Anvil, Astrazeneca,
  Aventis, Beyond
• Genomics, Biogen, Bristol Myers Squibb, Entelos,
  Gene
• Network Sciences, Genomatica, IBM Research,
  Ingenuity, ISB,
• KGI, Millennium Pharmaceuticals, Paradigm
  Genetics, Pfizer,
• Target Discovery and University of Pennsylvania.

17
Resource Sharing and Data ExchangeSBW and SBML

• Issues related to modern computing requirements
  for
• Systems Biology
• 1. Data Exchange between different Tools (SBML)
• 2. Access to Tool Functionality from other Tools
  (SBW)

18
SBML Systems Biology Markup Language
19
SBML Systems Biology Markup Language

• XML based Standard
• Simple Compartments (well stirred reactor)
• Internal/External Species
• Reaction Schemes
• Global Parameters
• Arbitrary Rate Laws
• DAEs (ODE Algebraic functions, Constraints)
• Physical Units/Model Notes
• Annotation extension capability

20
SBML Systems Biology Markup Language

• What is XML?
• XML stands for EXtensible Markup Language
• XML is a markup language much like HTML
• XML was designed to describe data
• XML tags are not predefined in XML.
• You must define your own tags

21
SBML Systems Biology Markup Language

• What is XML?
• XML does not DO anything
• XML was not designed to DO anything in
  particular.
• XML is created to structure, store and to send
  information.

22
SBML Systems Biology Markup Language

• What is XML?
• lt?xml version"1.0" ?gt
• ltnotegt lttogt Eloi lt/togt ltfromgt Morlock
  lt/fromgt ltheadinggt Reminder lt/headinggt
  ltbodygt I want to eat you lt/bodygt
  lt/notegt

23
SBML Systems Biology Markup Language

• XML has a hierarchical structure
• ltrootgt ltchildgt ltsubchildgt.....lt/subchil
  dgt lt/childgtlt/rootgt
• Each node can also have optional attributes, eg
  ltchild name johngt

24
SBML Example

• lt?xml version"1.0" encoding"UTF-8"?gtlt!--
  Created by XMLPrettyPrinter on 11/14/2002
  --gtltsbml level "1" version "1" xmlns
  "http//www.sbml.org/sbml/level1"gt lt!--
  --gt lt!-- Model Starts Here --gt
  lt!-- --gt ltmodel name
  "untitled"gt ltlistOfCompartmentsgt
  ltcompartment name "uVol" volume "1"/gt
  lt/listOfCompartmentsgt ltlistOfSpeciesgt
  ltspecie boundaryCondition "false"
  compartment "uVol" initialAmount "0" name
  "Node0"/gt ltspecie boundaryCondition
  "false" compartment "uVol" initialAmount
  "0" name "Node1"/gt
  ltspecie boundaryCondition "false"
  compartment "uVol" initialAmount "0" name
  "Node2"/gt lt/listOfSpeciesgt
• ltlistOfReactionsgt ltreaction name
  "J0" reversible "false"gt
  ltlistOfReactantsgt ltspecieReference
  specie "Node0" stoichiometry "1"/gt
  lt/listOfReactantsgt ltlistOfProductsgt
  ltspecieReference specie "Node1"
  stoichiometry "1"/gt
  lt/listOfProductsgt ltkineticLaw formula
  "v"gt lt/kineticLawgt
  lt/reactiongt
• ltreaction name "J1" reversible
  "false"gt ltlistOfReactantsgt
  ltspecieReference specie "Node1"
  stoichiometry "1"/gt
  lt/listOfReactantsgt ltlistOfProductsgt
  ltspecieReference specie "Node2"
  stoichiometry "1"/gt
  lt/listOfProductsgt ltkineticLaw formula
  "v"gt lt/kineticLawgt
  lt/reactiongt lt/listOfReactionsgt
  lt/modelgtlt/sbmlgt
• but
• of XML is therefore a compromise but one worth
  making so long as we understand and appreciate
  its limitations.

25
Other Related Efforts - CellML

• CellML is a more comprehensive attempt at
  developing an
• exchange standard, also defined in terms of XML.
• However, it is much more complex and the
  designers of CellML
• have not provided software support in the form of
  tools and
• software libraries.

26
Fitting Data to a Model
1. A model will most likely contain many
parameters whose values we will be very unsure
off. 2. These parameters can be made more
precise through a process of fitting experimental
data to the model. 3. This involves using
optimization algorithms which attempt to minimize
the difference between the model and the data.
27
Fitting Data to a Model
Data
Model
Optimization
Fitted Model
28
Fitting Data to a Model

• Questions
• How much data does one need to estimate
• the parameters of the model?
• 2. How can we obtain confidence in the parameter
• estimates we get?

29
Fitting Data to a Model
Common optimization methods include Simplex
Method - deterministic Genetic Algorithms -
stochastic
30
Simplex

• Simplex
• Evaluate the objective function at N1 points
  (simplex).
• Replace the worst point by reflecting it through
  the median of the remaining N points. We use a
  series of reflections, expansions and
  contractions to reach the minium.
• The simplex crawls over the objective function
  surface, untill it finds the minimum. The method
  is robust anbd does not involve computation of
  derivatives.

http//esd.lbl.gov/ITOUGH2/Minimization/minalg.htm
l
31
Simulated Annealing
Simulated Annealing The principle of this
method is that slowly cooled systems are able to
find their ground state. At a finite temperature
the system can jump to a higher energy state with
a certain probability. The method can thus escape
from local minima. It requires temperature
scheduling and is computationally intensive.
http//esd.lbl.gov/ITOUGH2/Minimization/minalg.htm
l
32
Genetic Algorithm
Genetic Algorithms Generate a random
population à select Parents using fitness
criteria à cross over and generate Children à
Mutate à Create new generation. Population
evolves à choose the fittest member. Evolve
several generations.
http//www-mctrans.ce.ufl.edu/featured/TRANSYT-7F/
Release9/Genetic.htm
33
Levenberg-Marquardt
Levenberg-Marquardt (LM) This method uses a
steepest descent type approach initially followed
by a quadratic decent. As the minimum is
approached, use the quadratic approximation. The
method automatically weighs these two methods.
Converges quickly, involves computation of
derivatives.
http//esd.lbl.gov/ITOUGH2/Minimization/minalg.htm
l