Transcript Document

Monday + Tuesday summary
By Frederic, Martijn, Hans
All Yellow slides have been seen/OK’d
by participants
Setting the stage for multiscale modelling:
From defunct molecules to cancer prognosis
• Heterogeneity between tumors is a key feature; more and
better tools are needed to diagnose the heterogeneity
between patients. Make a reference model and personalize
this on the basis of patient specific data
• Heterogeneity at the single gene level is much stronger
than at the phenotypic level
• A cancer phenotype, and its persistence is selected for. The
selection product thereby depends on the surrounding
tissue, hence the host, which is not even constant.
• The most important function may be energy, or energy,
carbon and nitrogen, the essentiality of which a cell cannot
mutate against.
• Selection of cells within a tumor also depends on
spatial heterogeneity and selection pressure
changes with time. Samples from a tumor are
often not representative.
• Tumor diversity comes from stochasticity at the
levels of mutations, low-molecule numbers
(transcription-translation), bistability, epigenetics.
• DNA changes are more persistent, unless selected
against during tumorigenesis. Some epigenetics
and metabolic states (glycogen) are semipersistent.
• Neither from-molecule-analysis will be effective
by itself, nor from-phenotype-analysis.
Multiscale modelling will need to bridge the two.
• Pathology should be made more quantitative,
linked up with models, and molecular
information should be weighed together with the
imaging data. This may require reclassification of
diseases.
• Therapy decisions and their effects should be
followed up for later testing of our models.
Multiscale modeling
• Subcellular level by ODE model. Agent based (1 cell=1 agent)
model for tissue level. Oxygen distribution through PDEs. Blood
flow important
• Take the perspective of only a few substances mediating
between the scales (e.g. intracellular and extracellular).
• Approach metabolism (mass transfer) distinctly from signalling.
• Model granularity depends on question asked
• Model reduction is important, but
• fully detailed model would be more useful than experimental
reality because the former can be interrogated more easily in
terms of predicting effects of therapy
Oxygenation/ROS
• Oxygen gradient depends on respiration which in
turn determines oxygen and glucose levels.
• Tumor cells can overcome many challenges
through adaptation except the energy issue
• Make models that specify the selective pressures;
mutation rates do not matter, mutation fixation
vis-a-vis the cells’ environment, does.
Tumor and bystanders
• 98% of metastases can be classified in terms of origin
• This may imply that the future metastatic capability
can be predicted from information in the primary
tumor. This could empower individualized therapy.
This constitutes a program of research of modelling to
predict metastatic potential. Problem: knowledge of
metastases is limited due to focus on primary tumors.
• Decide what to model: model functionally (e.g.
motility to understand metastasis, or metabolism for
tumor-cell survival), from there tumor anatomy,
pathways, molecules.
Modeling signaling pathways
• static vs dynamic (transient, sustained
responses)
• feedback circuitry
• link to metabolic network; e.g. nutrientbinding: metabolites can bind GPCRs; insulin
signaling
• cell context dependency
Genome-scale metabolic modeling
• Technical motivation:
- cell lines may not describe the tumor
- metabolic data (fluxes) hard to measure
- transcriptomic data readily available
• Mapping between scales; gene expression <> flux
scale. Somewhat predictive.
• because of network functional organization
(stoichiometric constraints) cells have to choose
between proliferation and consolidation (ROS
protection)
• higher growth rate correlates with longer survival
Molecular dynamics
• Talin two-state modelling/membrane
interactions shold be possible
• It is unclear at this stage whether a reduced
number of conformations (showing two
states) emerges.
How to bridge pathway level with cell
level (and up)?
•
Transparent black-box models {Consider the intracellular networks in high detail (150 000 types of
molecule) in transparent black box with a limited number of inputs and outputs (e.g. 70 of each);
Input/output to and from cells is limited (<150 species?); Model the intracellular as how all
molecules affect the transfer functions that lead from inputs to outputs.}. We must be able to
define a limited number of parameters that can be accurately measured (e.g. concentration of
metabolites homogeneous in cells) and make simplifications to be able to answer this problem.
•
Computational research agenda to test coarse graining strategy { Show at MD level that proteins
essentially live in a small number of conformation-areas. Then model proteins as existing in small
number of states., each with distinct activity. Model pathway in terms of activities. Show that the
heterogeneity stemming form the above converges to limited heterogeneity of pathway: treat next
level in terms of limited number of pathways. Show that this leads to limited number of cell states
that are frequent. Build tissue in terms of these. Etcetera; this all computational, although in
parallel experimental validation will be useful. Issue is how to transfer parameters between levels}
•
Robustness may (or may not), alleviate heterogeneity/stochasticity problems at higher scales. The
extent to which complexity can be reduced remains to be shown . Some models may already
oversimplify Life. . {Completely detailed models (Markus Covert, Cell 150, 2012) versus
understandable models (Palsson); Yet biology may be so complex that models need to be complex?
}
The call for proposals this would
define
• Develop multiscale models that predict
metastatic activity/success on the basis of excised
tumor material, as well as utility of and type of
personalized anti-metastatic therapy
• Develop multiscale models of tumor
energetics/survival inclusive of oxygen, glucose,
pH gradients, that take tumor cell evolutionary
success and intratumor heterogeneity into
account and suggest new personalized networkbased drug targets.
• Perhaps set this up competitively:
Proposal of modelling competition
• Modelling competition? Like Asilomar’s CASP; homology/threading
modelling
• DREAM project exists in systems biology: top down modelling; the new
proposal could be for bottom-up modelling.
• NIH: Simon Kasif (BU, George Church); funds experiments that test model
predictions.
• Specific data set plus questions plus modelling types need to be defined
(because existing modelling methods are complementary rather than
parallel)
• Question could be applied (e.g. medical) or fundamentally scientific (e.g.
Warburg effect). Question should not be too complex/complicated
• But this may not be enough because we cannot define what an
appropriate dataset would be. May not be multiscale.
• iGEM may be a good analogy; BUT this is more difficult to model than MD
is. Questions may need to be limited in complexity. May be too
expensive.
• Link with hospitals? But should then be professional.