Modeling Complex Biological Systems: Examples in Computational Cell Biology, Computational Neuroscie

1
Modeling Complex Biological Systems Examples
in Computational Cell Biology, Computational
Neuroscience and applications to Computer Science
Hugues BERRY
Project-Team ALCHEMY, INRIA Saclay-Île de France
Research Centre and LRI, UMR 8623 CNRS -
University Orsay Paris-Sud
2
Life at the molecular level is disordered

• Static view gene regulation networks

Yeast transcription netw.
Milo et al. (2002)
regulating
regulated
The average connectivity
exists but its fluctuations
3
Life at the molecular level is disordered

• Dynamic view single cell kinetics

Elowitz et al. (2002)
Intracellular biochemical reactions are
intrinsically stochastic
4
Life at the cellular level can be disordered

• Example neurons

Druckmann et al. (2007)

• Response of the same neuron to the same input in
  vitro

• Connectivity network of cortical areas (cat)

Sporns . Zwi (2004)

• Membrane potential of a cortical neuron in vivo

Lampl et al. (1999)
5
But order appears at larger length scales

• Self-organization of these noisy/disordered
  behaviors yields coherent global behaviors
• Embryogenesis
• Physiological rhythms
• Brain functions

6
Motivations

• How do biological systems cope with their
  constitutive noise / disorder?
• Do biological systems take advantage of this ?
• Can we exploit this for technology-oriented goals?

• Need modeling / simulation approaches
• Computational cell biology
• Computational neuroscience

• Applications to computer science
• Bio-inspired computation

Ask not only what CS can do for biology, ask as
well what biology can do for CS
7
Outline

• Coping with noise
• Memory - plasticity in single neurons
• Exploiting disorder
• Geometrical disorder in cell biochemistry
• Hebbian learning in chaotic neural networks
• Applications to computer science
• Amorphous cortex-inspired image recognition
  systems
• Engineering cells synthetic biology
• Conclusion
• Acknowledgements

8
1. Coping with noise
9
Plasticity and memory in single neurons
with B. Delord, S. Genet and E. Guigon

• Neurons continuously modify their membrane
  properties in response to their inputs
  information storage
• Information storage in neurons modification of
  the synaptic weight
• includes plasticity (modification) and memory
  (maintenance)
• can be multivalued (vs binary)
• doesnt last forever (sec to years)
• All these aspects can be accounted for by a
  simple, realistic biochemical module Delord et
  al. (2007) PLoS Comp Biol

10
Realistic conditions within dendrites

• Noise (low copy nbrs) sets memory logic
  (capacity) but information storage is robust

• Dendrite size (0.05 µm3) proteins are in low
  copy numbers

Delord et al. (2007) PLoS Comp Biol
11
2. Exploiting disorder
12
Geometrical disorder in cell biochemistry

• Macromolecular crowding inside cells is huge
  Verkman (2002)
• As a consequence, diffusion is anomalous
  (subdiffusion)
• for experimental measurements see e.g.
• Weiss et al. (2004) or Guigas et al. (2007)
• Subdiffusion is not a well-mixing process
• Spatial fluctuations not efficiently damped

D. discoideum with cryoelectron tomography
Madalia et al., 2002
13
Geometrical disorder in cell biochemistry

• Example 2D enzyme kinetics with immobile
  obstacles

Berry (2002) Biophys J
14
Take-home Message

• Molecular crowding in cells expected to
• modify thermodynamics (equilibrium, binding
  constants) Minton (2001)
• modify the kinetics of bacterial gene regulation
  (target finding times for transcription factors)
  Golding Cox (2006) Guigas Weiss (2008)
• change mass-action laws into power-law decays
  Berry (2002) Biophys J
• amplify spatial fluctuations of species
  concentration Berry (2002) Biophys J
• Geometrical disorder could be a process to
  generate non trivial spatial organization within
  cells

15
Perspectives Aging in bacteria
with A. Lindner and F. Taddei

• Aging (? growth rate with time) exists in
  bacteria Stewart et al. (2005) PLoS Biol
• Related to the aggregation of a chaperon protein
  IbpA Lindner et al. (2008) PNAS
• Nontrivial aggregation pattern

• Aim Uncover the molecular mechanisms
  responsible for this spatial pattern
  (macromolecular crowding, membrane interactions)
• Ex Pure 3D diffusion-aggregation process (no
  obstacles)

16
Complex dynamics in cortical networks
with B. Siri (PhD), B. Cessac, B. Delord and M.
Quoy

• is mainly due to complex connectivity
• Synaptic plasticity yields a mutual coupling
  between neuron activity and synaptic network
  topology
• This mutual coupling is still poorly understood
  in the case of complex dynamics

neuron state
connection weight
17
Evolution of the dynamics complexity
Siri et al. (2007) J Physiol Siri et al. (2008)
Neural Comput
Systematic decay of the dynamics complexity

• Sensitivity to input pattern is max at the edge
  of chaos
• Chaos rich reservoir of possible behaviors
  (diversity) but no coding
• Hebbian learning selects elements of this
  reservoir to encode input

18
Perspectives

• Redistibution of the weights over the network
  still random but increased correlations
  Small-world topology Berry Quoy (2006)
  Adaptive Behavior Siri et al. (2007) J Physiol
• observed in all biological neural networks to
  date
• are some topological classes (small-world,
  scale-free, random, ordered) more adapted than
  others for neural networks tasks?
• use evolutionary strategies for the topology
  (with M. Schoenauer, F. Jiang PhD thesis)

19
3. Applications to computer science
20
Evolution trends in computer architectures

• Increasing computing units, disorder (nanotech
  self-assembly), decentralization, probability of
  faults/defects
• How to design, organize, program such systems?
• Take inspiration from complex biological systems
• Self-organization local interactions ? coherent
  global behaviors
• Biocomputation use space rather than individual
  speed
• Disordered structures robustness to
  faults/defects
• Compute with biological hardware (wetware)?

21
Cortex-inspired computing
with E. de Labareyre and O. Temam

• Sanity check Complex visual object recognition
• Good biological understanding data
• Human primates outperform best machine systems
  on several aspects

Riesenhuber Poggio (1999)
22
Computing with biological cells?

• First possibility
• Use neuron-chips interfaces (MEA) and
  input/output conditioning
• Develop dedicated adaptive control methods

Baruchi BenJacob (2007)
Randomly-assembled computer Lawson Wolpert
(2006)
23
Perspectives
with H. de Jong, A. Lindner, G. Batt, J.L. Gouzé,
H. Geiselman et al.

• Controlling bacterial growth
• growth rate for bacteria evolution fitness
• but growth rates in a population show extensive
  variability
• this variability will have to be controlled in
  computing applications
• A systems biology and synthetic biology approach
  to
• decipher the endogenous molecular networks for
  growth and aging
• control average growth rate and variability at
  will through synthetic rewiring of the network
  and adding synthetic genetic controllers
• Co-funded by INRIA and INSERM

A
B
aging
24
Conclusion

• Why are biological systems so complex?
• Highly multi-scale molecules, organelles, cells,
  organs, bodies, populations, ecosystem
• Bottom-up top-down causality
• High heterogeneity of the elements, at each scale
• Disordered structures based on self-organization
• Evolution could have turned constraints into
  opportunities
• e.g. noise diversity
• Understanding this is a key issue that will need
  help
• Mathematics, statistical physics, computer
  science
• Multidisciplinary approaches
• Promises of great expectations
• Understand what is specific about life
• New technological developments, beyond classical
  biotech

25
Acknowledgement

• Students
• G. Caron-Lormier (Master 2001-2002), now postdoc
  in Harpenden, UK.
• D. Pellenc (PhD, 2002-2005), now postdoc in
  Reading, UK
• B. Siri (PhD, 2005 - )
• F. Jiang (PhD, 2006 - )
• E. de Labareyre (Master 2008)
• Grants
• ANR ASTICO (Learning in complex biological
  systems), 2005-2008
• INRIA ARC AMYBIA (Aggregating Myriads of
  Bioinspired Agents), 2008 - 2009
• INRIA ARC MACACC (Modelling cortical activity
  and analysis of the cerebral code), 2008 - 2009
• CNRS PEPS MARTINE (Multifractal Analysis to
  Resolve information Transfer In NEural networks),
  2008

26
Acknowledgement

• Collaborations
• CEA List D. Gracia-Pérez
• INLN, Nice B. Cessac
• INRIA G. Batt, N. Fates, B. Girau, M.
  Schoenauer, O. Temam
• INRIA Neurospin B. Thirion
• INSERM Cochin A. Lindner, F. Taddei
• INSERM Jussieu B. Delord, S. Genet, E. Guigon,
  J. Naudé
• Thales R T S. Yehia
• Univ. Cergy-Pontoise F. Germinet, M. Quoy
• Univ. Evry - Genopole J.L. Giavitto, O. Michel
• Univ. Lyon CNRS H. Paugam-Moisy

27
Further information / papershttp//www-rocq.in
ria.fr/hberry
28
Life at the molecular level is disordered

• Dynamic view single cell kinetics

Elowitz Leibler (2000)
29
Performance dynamics of µprocessors
with D. Gracia-Pérez and O. Temam

• Current microprocessors are regular finely
  engineered
• Yet the dynamics of their performance is already
  complex
• Analyze it using tools from complex biological
  systems

Berry et al. (2006) CHAOS