Transcript November 17, 2015 Team 9 (Sarojini Attili, Kimberly Taylor)

Sarojini Attili
Kimberly Taylor
Sample D
2
Sample A
Sample D
Sample B
Sample E
Sample C
3
Sample A
Sample D
Sample B
Sample E
Sample C
4
Sample A
Sample D
Sample B
Sample E
Sample C
5
Sample A
Sample D
Sample B
Sample E
Sample C
Computational
model of the human
body that integrates
all the different
whole cell models
Computational
model of the
pathogen
6
Sample A
Sample D
Sample B
Sample E
Sample C
Details of
mutation/s and
Phenotypic data
Pathogen
specific data
Computational
model of the human
body that integrates
all the different
whole cell models
Computational
model of the
pathogen
7
Sample A
Sample D
Sample B
Sample E
Sample C
Details of
mutation/s and
Phenotypic data
Pathogen
specific data
Computational
model of the human
body that integrates
all the different
whole cell models
Computational
model of the
pathogen
8
Sample A
Sample D
Sample B
Sample E
Sample C
Whole cell modeling
for personalized
medicine
Details of
mutation/s and
Phenotypic data
Pathogen
specific data
Computational
model of the human
body that integrates
all the different
whole cell models
Computational
model of the
pathogen
9
Modeling Biological Systems
• Significant task of systems biology and mathematical
biology
• Computational systems biology aims to develop and use
 Algorithms
 Data structures
 Visualization
 Communication tools
• Goal: Perform computer modeling of biological systems.
• It involves the use of computer simulations of biological
systems to both analyze and visualize the complex
connections of these cellular processes.
10
Whole cell modeling
 Developing a complete model
for a specific cell that includes
all pathways, processes and
functionality.
 The whole cell model explains
the entire lifecycle of the cell.
 The authors of this paper have
modeled the life cycle of
individual Mycoplasma
genitalium cells.
 Whole cell modeling although
challenging, is very important
for the future of medicine.
11
Why whole cell modeling?
• Whole cell models comprehensively predict how
•
•
•
•
phenotypes emerge from genotypes.
Whole-cell modeling could enable rational bioengineering
and precision medicine.
Whole cell models could also enable clinicians to
individualize therapy.
Combined with genome synthesis and transplantation,
whole-cell models could enable bioengineers to produce
biofuels.
Overall, whole-cell models could be powerful scientific
tools.
12
History of whole cell modeling
• Beginning in the late 1970s, researchers began modeling
cell physiology, primarily using ordinary differential
equations, creating increasingly detailed models over the
next three decades.
• Later on, other groups introduced frameworks that require
fewer parameters than ODE systems including constraintbased and Boolean methods.
• Combining these approaches the authors of this paper
developed a hybrid methodology to model the life cycle of
individual Mycoplasma genitalium cells – Individual
biological processes were modeled, each with its own
mathematical representation and individual outputs were
merged to compute the overall state of the cell.
13
Some existing computational models
 Cardiac system models – The first model developed for
the heart was the Hodgkin–Huxley model, today we
have more sophisticated models
 Computational models for different cellular processes
or parts of the cell such as:
 The dynamics of Ca2+ wave propagation during
xenopus oocyte maturation
 Dynamics of calcium sparks and calcium leak in the
heart
 Metabolic model of the mitochondrion
14
Outline
 Core principles of whole-cell modeling
 Model construction process outline
 Example of a whole cell model
 Challenges to achieving complete
models
15
The principles of whole-cell
modeling
16
Single cellularity:
Whole cell models should
represent individual cells.
Single cell models can account
for how temporal dynamics and
stochastic variation affect
behavior. Single cells are also
tractable because they are
independent and directly result
from molecular biochemistry.
17
Functional closure:
Behavior is determined by interacting pathways and genes. Consequently,
whole-cell models should represent every known cellular and gene function.
Models which represent every known function are powerful tools. For example,
genome-scale metabolic models which represent every known metabolic
reaction and enzyme have been used to identify missing reactions and
enzymes.
18
Molecular closure:
Whole-cell models should represent the cell and its environment as a
closed system. Models should explicitly account for exchanges among
pathways and the environment and not have arbitrary sources and
sinks.
Temporal closure:
Whole-cell models should also represent the entire cell cycle. This
ensures that models account for how cells regulate pathways in time to
coordinate their life cycle. For example, models should account for how
the dynamics of DNA replication affect dNTP concentrations and
metabolism.
19
Biophysics:
In addition, whole-cell models should represent cellular biochemistry
and biophysics, including mass conservation, thermodynamics, and
spatial organization. Some of the methods capable of representing
cellular biophysics are - molecular dynamics, Brownian dynamics, latticebased models. Below are some representations of space:
20
Dynamics:
S+E
In particular, whole-cell
models should be constructed
from differential descriptions
of molecular biochemistry and
predict the emergence of
cellular-scale dynamics.
Emergent dynamics are
valuable opportunities for
experimental validation and
discovery.
k1
k-1
C
k2
P+E
ds
 k 1c  k1se
dt
de
(k 1  k 2 )c  k1se
dt
dc
 k1se  (k 2  k1 )c
dt
dp
 k2c
dt
21
Stochasticity:
Furthermore, whole-cell models should be discrete and
stochastic. Stochastic models naturally predict the
emergence of cellular variation. For example, stochastic
models can account for how stochastic transcription
initiation creates variation in gene expression and growth.
This variation is another valuable opportunity for
experimental validation.
Species specificity:
Whole-cell models must be evaluated by comparison to
experimental data. Consequently, whole-cell models
should represent specific genomes. This constrains the
space of training data.
22
Parsimony:
Despite the explosion in experimental data, limited data is
available. For example, there is little data about non-coding
RNA. Consequently, models should be parsimonious. This
minimizes the need to identify unmeasured parameters.
Modularity:
Like other large engineered systems, whole-cell models are
best developed by combining multiple pathway submodels.
This enables submodels to be developed and tested
independently by different investigators using different
representations.
23
Reproducibility:
Finally, whole-cell models should be transparent, well-annotated, and
reproducible. Researchers should be able to reproduce models from
their primary sources, as well as reproduce simulations using multiple
simulators. Models should also be described using transparent
languages, this is essential for collaborative modeling. Example: SBML
24
Model construction
 Experimental





curation
Mathematical
formulation
Submodel
integration
Parameter
estimation
Model refinement
and validation
Visualization and
analysis
25
Experimental curation:
The first step to constructing a model is to choose an
organism and assess the feasibility of modeling it by
assembling the available experimental knowledge.
Some examples of experimental data sources include organism database tools such as Pathway Tools,
WholeCellKB, BioMart and Intermine.
26
Mathematical formulation:
• A mathematical description of how cells evolve over time must be constructed.
• Describing the cell as thoroughly as possible using existing knowledge avoids
unknown parameters and expensive computations.
• One model can be used for many scientific questions.
• Individual submodels must be implemented and/or constructed from
experimental data.
• Databases like BioModels and CellML contain many existing pathway models.
• Rule-based modeling is a powerful and scalable approach for assembling
genome-scale models.
Some of the tools that can be used to generate mathematical models include:
• BioNetGen
• BioUML
• CellDesigner
• CobraPy
• COPASI
• E-Cell
• iBioSim
27
Submodel Integration:
• The individual submodels that were developed as part of the
mathematical formulation must be combined.
• Homogeneous submodels can be merged analytically.
• Heterogeneous submodels must be combined in a multistep approach;
hybrid simulators have been developed recently which are capable of
integrating heterogeneous submodels.
28
Parameter Estimation:
Once the model's structure has been implemented, the model's
parameters must be identified by matching the model's predictions to
experimental data.
29
Model refinement
and validation:
Next, an important step to
constructing a whole-cell model
is to iteratively evaluate the
model's predictions and refine
the model.
Predicted phenotypes of genetic
perturbations should be
evaluated.
Methods used to automate model
refinement:
 Robotic and microfluidic
experimentation
 Computational gap filling
30
Visualization
and analysis:
The last step in whole-cell
modeling is to simulate the
model, analyze simulation results
to construct new hypotheses, and
conduct experiments to test those
hypotheses.
31
Modelling example
 Karr, JR et al. (2012) A whole-cell computational model
predicts phenotype from genotype. Cell 150: 389-401
 One of the most complete and rigorous models to date
32
Modelling of Mycoplasma
genitalium
 Small bacterium isolated from urethra in 1980
 Causes urethritis (inflammation of the urethra) in
both men and women
 Implicated in HIV transmission
 Genome is single circular dsDNA with 525 genes
33
Goals of M. genitalium modelling

Describe the complete life cycle of a single cell on
level of single molecules
 Replicate function of every known gene product
 Predict multiple cellular behaviors, including
macromolecular synthesis and the complete cell
cycle
Models were based on 900+ publications and 1900+
observed parameters
34
How did they do it?
 Identified 28 modules covering cellular functions
 All modules independently built, parameterized and
tested
35
How did they do it?
 Modules were assumed independent at time scales < 1 s
36
Model integration
37
Why modules?
• Module-based modelling proposed in 1999
– 668 citations in PubMed
• Model is based on experimental observations
1. Action potentials
2. Decision making in bacteriophage λ
38
Hartwell et al. (1999) From molecular to modular cell biology. Nature 402: C47-C52
Experimental validation
39
Exploring the cell cycle
40
Global energy analysis
41
Phenotype studies
 Disruption studies performed for all 525 genes
 284 essential genes and 117 non-essential
 Model allows prediction of phenotype from known
genotype
42
Challenges in whole-cell
modelling
 Macklin, DN et al. (2014) The future of whole-cell
modeling. Current Opinion in Biotechnology 28:111–115
43
Challenges in whole-cell
modeling
44
Experimental interrogation
 M. genitalium was chosen because of its small genome,
but there are relatively few papers on this pathogen
 Future research will focus on well-studied organisms
such as E. coli, S. cerevisiae or Mycoplasma
pneumoniae
 E. coli:
 347,790 articles in PubMed (as of 11/12/2015)
 4,288 protein-coding genes
45
Data curation
 Where and how do you store the data?
 Data for M. genitalium available at
http://www.wholecellkb.org/
 Automatic curation will be needed
 Human and machine involvement
 Separate database for each organism
 Question: how should the data be formatted?
46
WholeCellKB
47
http://www.wholecellkb.org/
WholeCellKB
http://www.wholecellkb.org/
48
Other considerations
 Mathematical models could use a Boolean “switch” for
communication between modules and sub-modules
 Cell behavior cannot violate physical laws
 Model must be consistent with biological phenotypes
49
Accelerated computation
 ~10 hours were required for each simulation of M.
genitalium
 Simulation of 525 single-gene disruptions required 5250
hours, or ~220 days
 In theory, a single simulation of E. coli would take 81
days!
50
Speeding up computation
 High performance parallel computing
 Custom hardware platforms
 Could this be an application for quantum computing?
51
Analysis and visualization
 Extensive analysis of raw data is needed to interpret
results
 Machine learning
 Dynamic systems analysis
 How do you visualize large data sets?
52
Model validation
 Feedback between model and experimental results
 M. genitalium work was more intuitive than rigorous
 Quantitative metrics must be developed
53
Collaboration and community
development
 Code base for the M. genitalium whole-cell model
released under MIT license
 Whole cell modelling will proceed faster with
collaboration between all researchers
 Will competing groups want to share their results before
publication?
 Is a uniform format needed?
54
Conclusions
 Whole-cell modelling has been shown to be successful
 Validated by experimental results
 Phenotype predicted from genotype
 Potential for study of cell processes that cannot be
addressed experimentally
 Improvements in analysis, visualization and technology
are needed
55
Future directions
 Non-bacterial cells
 Cells with large genomes
 With current technology, simulation of single human
cell (3 billion bp, ~25k genes) would take 476 hours or
~20 days!
 Cell-cell interactions
 Modelling of organs and organ systems
 Modelling of multicellar organisms
56
References
 Karr, JR et al. (2015) The principles of whole-cell
modeling. Current Opinion in Microbiology 27: 18-24
 http://www.sciencedirect.com/science/article/pii/S1369
527415000685
 Karr, JR et al. (2012) A whole-cell computational model
predicts phenotype from genotype. Cell 150: 389-401
 http://www.sciencedirect.com/science/article/pii/S0958
166914000251
 Macklin, DN et al. (2014) The future of whole-cell
modeling. Current Opinion in Biotechnology 28:111–115
 http://www.biomedcentral.com/1471-2105/14/253
57
References
 http://biologypop.com/the-evolution-of-the-cells/
 http://www.biomedcentral.com/1752




0509/2/72/figure/F4?highres=y
http://oreillyscienceart.com/figures-for-publication/
http://www.elveflow.com/microfluidic-tutorials/cellbiology-imaging-reviews-and-tutorials/microfluidic-forcell-biology/concepts-and-methodologies/
https://openi.nlm.nih.gov/detailedresult.php?img=322438
2_1752-0509-5-155-1&req=4
http://www.biomedcentral.com/1471-2105/14/253
https://www.google.com/imghp
58
Thank you!
59