ankursinha.in/blog

neuroscience/fedora/musings

Fri 14 August 2015

CNS 2015 - Workshops

Posted by ankur in Research (854 words, approximately a 4 minute read)


  • Share this post:

The notes have not been proofread. Please do your research before you pick anything from this post. It is meant to be a rough sketch of everything that I heard and noted at the conference. Since quite a bit of this is new to me, it is bound to be inaccurate, unspecific, and possibly even incorrectly quoted.

Dendritic growth and wiring workshop

  1. Hippocampus - theta oscillation and place cell firing -> directly proportional to locomotion speed.
  2. Place related firing of place cells is dendritic function.
  3. Can control the speed of locomotion by external stimulation -> reaches a plateau.
  4. CITE: Fuhrmann 2015
  5. Cerebellar synaptic units: bouton and synapses placements and distance distribution
    1. Double boutons - split and meet 2 neurons.
    2. Purkinje cells have access to many boutons - lots of boutons close by
    3. Not same as interneurons which have few boutons around
    4. Peter's rule of synapse formation.
    5. PF connectivity is constrained by volume exclusion.
  6. Deconstructing and reconstructing neuronal morphology - George Ascoli
    1. Biochemistry is very similar to muscles
    2. B(ranch)A(dvance/extend)R(etract)T(urn)
    3. CITE: Zhou et al. 2002
    4. Neuromorpho.org
    5. BOOK: Trees of the brain, roots of the mind
  7. Computational modelling of synaptic and dendritic plasticity in the hippocampus - Peter Jedlicka
    1. Neural activity -> synaptic plasticity -> synaptic efficiency -> neural activity ...
    2. Dentate Gyrus - (LTP + heterosynaptic STD) (homoeostatic mechanism)
    3. Model based on STDP and BCM
    4. STDP:
      1. LTP amplitude and :math: tau
      2. LTD amplitude and :math: tau
    5. BCM homoeostasis rule LTP/LTD amplitudes vary, proportional to integrated spike count (60 second).
    6. MPP and LPP are spontaneously active.
    7. Removing BCM homoeostasis/metaplasticity deteriorates performance
    8. If the mechanism is too quick, plasticity is hampered - why?
    9. Working on dendritic remodelling following lesions -> computational modelling
    10. Functional relevance of dendritic retraction
    11. Algorithms based on wiring optimisations
    12. BAPs are selectively stronger in the denervated dendritic layer
    13. At a dendritic morphology level
    14. Homoeostasis following dendritic retraction
    15. Smaller neurons fire more because their resistance is higher(?)
    16. Compensatory enhancement of the intrinsic excitability -> in spite of lower number of stimulated
    17. Excitability homoeostasis is present in all dendritic trees which undergo lengthening or shortening of their branches while keeping synaptic density constant
    18. Friedemann is working on homoeostasis and time scales
  8. Different inputs project on different parts of the dendritic tree
    1. Both BAPs and Calcium spikes are involved in synaptic plasticity.
    2. Inhibition can switch plasticity on and off - hypothesis
    3. BAP (sodium spike)
    4. An inhibitory neuron makes about 12 connections to a post synaptic neuron?
    5. Calcium spikes cause soma to burst
    6. This is all a GABA-a inhibition

Plasticity workshop

  1. Concepts
    1. Pathways are different for LTP and LTD and so on
    2. Synaptic plasticity
  2. Hebbian plasticity and STDP is unstable.
  3. Homoeostasis
    1. On synapses #. Local - homosynaptic #. Global - heterosynaptic
    2. Change firing properties
      1. Threshold
      2. Intrinsic excitability
    3. Modulation of plasticity rule (metaplasticity)
    4. Anti-hebbian
    5. Putative role of glial cells
    6. Synaptic scaling - slow (\(\tau_{home} = 1~day\))
    7. CITE: Kech 2013 - lesioning is followed by upscaling synapses also
    8. CITE: Zenke 2013 - metaplastic triplet
    9. All of them have the same behaviour - oscillations if the time scle of STDP is much larger than that homoeostatic mechanism.
    10. Functional significance of oscillations in weights? No answer yet.
  4. Synaptic plasticity and scaling their role in memory formation
    1. Can SPM explain dynamics of memory?
    2. Memory consolidation during sleep -> important
    3. Once an assembly (memory) has been formed, they can be consolidated by a completely unspecific stimulus
    4. Slow wave sleep is important for consolidation
    5. Recall by partial stimulus causes weakening of assembly because of imbalance of activity - recall acts as perturbation
    6. Similarly, learning a related, overlapping task also has the tendency to act as a perturbation.
  5. Modelling the dynamic interaction between hebbian and homoeostatic plasticity (synaptic scaling)
    1. Hebbian is fast
    2. Homoestatic is slow
    3. How can homoeostatic plasticity be powerful enough and slow enough at the same time?
    4. If homoeostasis is slow, it can result in oscillations
    5. Data from monocular deprivation
    6. TNF-alpha mediates homoeostatic plasticity
    7. Spines fluctuate - explain volume distribution in absence of neuronal activity
    8. Fluctuation is directly proportional to spine size
    9. CITE: Yasamatsu 2008
    10. Small spines tend to stay for longer periods
    11. CITE: Kasai et al 2010
    12. Spike fluctuation are responsible for synaptic normalisation (hypothesis!)
    13. At a population level spike volume exhibits bistability

Open source brain workshop

  1. NeuroML - XML for computational neuroscience
  2. neuroconstruct
  3. neuromorpho
  4. channelpedia
  5. Allen database
  6. pyNeuroML and jNeuroML
  7. Openworm
    1. Cell by cell 3D neurmechanical model
      1. elegans
    2. 302 neurons
    3. 956 cells
    4. Fully sequenced genome
    5. Current state:
      1. Ready issues are mentioned on GitHub main page like Travis build status
      2. Wormsim

Large scale modelling workshop

  1. Visual cortex in NEST - Sacha Van Albada
  2. Towards rewiring of full scale cortical networks - Rinke and Butz
    1. Synaptogenesis
      1. Directly proportional to f(distance)
      2. Direclty proportional to f(availability of neurites)
      3. Time per axon: \(O(n)\)
      4. Time total \(O(n^2)\)
    2. Deletion:
      1. Time per axon: \(O(1)\)
      2. Time total \(O(n^2)\)
      3. For increasing number of neurons, the runtime increase drastically. (\(10^3 \rightarrow 100ms, 10^4 \rightarrow 10s, 10^5 \rightarrow 16m, 10^6 \rightarrow 27 hours\))
      4. The probability calculation for 1 neurite takes 100ms
      5. Optimisation method is inspired by the Barnes-Hut n-body method.
        1. Formation
          1. Tree construction - \(O(n log n)\)
          2. Time per axon - \(O(log n^2n)\)
          3. Total: \(O(nlog^2n)\)
        2. Deletion
          1. Time per axon - \(O(1)\)
          2. Total: \(O(n)\)
        3. Runtime (\(10^3 \rightarrow 10ms, 10^4 \rightarrow 200ms, 10^5 \rightarrow 3s, 10^6 \rightarrow 40s, 10^7 \rightarrow 10min\))
        4. Further parallelisation will reduce times even further
        5. Current and future:
          1. Validate results
          2. Parallelise

  • Share this post:

Comments