Uni-Logo
You are here: Home BCCN Research A5: Structure-function analysis using statistical relational learning and link mining
Document Actions

A5: Structure-function analysis using statistical relational learning and link mining

A5: Structure-function analysis using statistical relational learning and link mining

Stefan RotterB, Ulrich EgertP, Michael FrotscherH,O and Luc De RaedtK

 

K = Machine Learning Lab, Institute of Informatics;
H = Anatomical Institute,
O = Center for Neuroscience,
B = Bernstein Center Freiburg
P = Biomicrotechnology

Scientific background

Neuronal networks are, at a structural level, similar to social networks, citation graphs, the Internet, or networks of biochemical reactions: they are large graphs, linking many interacting elements to each other. Previous approaches to infer dynamic network coupling and its relation to anatomical connectivity from neuronal activity dynamics relied mostly on correlation analyses. A novel class of analysis tools, aimed at discovering regularities and structure in data from networks, is being developed in the fields of machine learning and data mining. These are known as "statistical relational learning", "link mining and discovery", "multi-relational data mining", and "probabilistic inductive logic programming". They have been successfully applied to a wide range of problems in the life sciences, the Internet and social networks, but not yet to discover structure-function relationships (SFR) in biological neuronal networks.

Objectives

The relation between structure and function is a central theme in neurobiology. Many questions, however, have not yet found definite answers: To what extent does network structure determine activity dynamics and function? Can the same structure support different dynamics and functions, depending on the context they are used in? How robust is the function when the structure develops, adapts, degrades, or otherwise changes? Do structural changes imply graceful changes in function? Does "usage" of a network change its structure? The objective of this project is to automatically discover SFRs and their changes in plastic biological and model neuronal networks, based on dynamic activity data as gathered in projects A2, A3, B1, B2 and B4. Existing tools from statistical relational learning and link mining will be adapted, and novel tools will be developed. Both intra- and inter-SFRs are targeted, with intra-SFRs identifying regularities within a single network, and inter-SFRs identifying common structures shared by multiple networks.


List of project-related publications

  1. T. Gürel, S. Rotter, and U. Egert. Functional identification of biological neural networks using reservoir adaptation for point processes. J Comput Neurosci:10.1007/s10827-009-0176-0, 2009.
  2. Gürel T, De Raedt L, Rotter S. Ranking the Neurons for Mining Structure-Activity Relationships of Biological Neural Networks: NeuronRank. Neurocomputing 70, 1897-1901, 2007.
  3. Gürel T, De Raedt L, Rotter S. Mining Structure-Activity Relations in Biological Neural Networks using NeuronRank. Perspectives of Neural-Symbolic Integration. Chapter 3, 47–63. Springer-Series in Computational Intelligence, 2007.
  4. Gürel T, Egert U, Kandler S, De Raedt L, Rotter S. Predicting Spike Activity in Neuronal Cultures. In Proceedings of the IEEE International Joint Conference on Neural Networks IJCNN-2007, Orlando, FL, August 2007.

 

Posters:

  1. Gürel T, De Raedt L, Rotter S. NeuronRank for Mining Structure-Activity Relationships of Biological Neural Networks. Fifteenth Annual Computational Neuroscience Meeting CNS*2006, Edinburgh, UK, 2006.
  2. Gürel T, Kersting K, Kandler S, Egert U, Rotter S, De Raedt L. Learning the Functional Connectivity in Neuronal Cultures. 2nd Bernstein Symposium for Computational Neuroscience, Berlin, October 2006.
  3. Gürel T, Kersting K, Kandler S, Egert U, Rotter S, De Raedt L. Learning the Functional Connectivity in Neuronal Cultures. 31st Göttingen Neurobiology Conference, March 2007.
  4. Gürel T, Egert U, Kandler S, De Raedt L, Rotter S. Predicting Spike Activity in Neuronal Cultures. Sixteenth Annual Computational Neuroscience Meeting CNS 2007, Toronto, Canada, July 2007.
  5. Gürel T, Okujeni S, Weihberger O, Rotter S, Egert U. Burst Clustering and Prediction in Neuronal Cultures. Frontiers in Computational Neuroscience. Conference Abstract: 4th Bernstein Symposium for Computational Neuroscience, Munich, October 2008.
  6. Jarvis S, Gürel T, Rotter S and Egert U (2008). Effects of network composition against network stability and temporal resolution in recurrent neural networks: identifying criteria for network bursting. Frontiers in Computational Neuroscience. Conference Proceedings: Bernstein Symposium 2008. doi: 10.3389/conf.neuro.10.2008.01.109
  7. O. Weihberger, S. Okujeni, T. Gürel, and U. Egert. State dependent I/O gain and interaction with ongoing activity in cortical networks in vitro. S20-2, 2009.
  8. T. Gürel, S. Rotter, and U. Egert. Reservoir computing methods for functional identification of biological networks. 18th Annual Computational Neuroscience Meeting CNS 2009.
  9. S. Jarvis, S. Rotter, and U. Egert. Clustered network topology and noise directly influence quasistable bursting behaviour. Proceedings of the Göttingen Meeting of the German Neuroscience Society 2009T26-12B, 2009.
  10. S. Jarvis, S. Rotter, and U. Egert. Effect of structural network composition on network stability and dynamics. BCCN Conference 2009, Frankfurt

 

Conference Talks:

  1. Gürel T, Egert U, Kandler S, De Raedt L, Rotter S. Predicting Spontaneous Activity and Modeling Input-Output Relations in Neuronal Networks. 3rd Bernstein Symposium for Computational Neuroscience, Göttingen, September 2007.
  2. Gürel T, Egert U, De Raedt L, Rotter S. Modeling Spontaneous Activity and Stimulus-Response Relations of Biological Neural Networks. Dagstuhl Seminar 08041, “Recurrent Neural Networks- Models, Capacities, and Applications”, January 2008.
     
Personal tools