Maximum Likelihood Estimation based Connection Reconstruction (MLECR)
Understanding the synaptic connections and organization of neural networks is a key step in understanding the dynamics of the brain. A large number of experimental techniques enable us to reconstruct these connections, but these methods are limited and only allow us to reconstruct networks of up to a dozen of cells. An alternative method is to record the activity of neurons and attempt to fit a model to the collected spikes. There are two challenges in this approach: simple models are often unable to resolve ambiguities and complex models are computationally very expensive.
In the MLECR project, issues of both explanatory power and computational tractability are handled using a method for reconstructing the connections in large scale networks of N > 1000 neurons as demonstrated in Zaytsev et al. (2015). Assuming a simple neuron model, a maximum likelihood estimation is made for the probability that the observed spikes are generated by a model with parameters W. It is possible to traverse the gradient of the likelihood estimate and calculate the optimal set of parameters W* for the model. For simulated neural networks, the MLECR method achieves good performance for realistic models using realistic amounts of input data; it is also capable of reconstructing embedded dynamic network structures like synfire chains.
Our Contribution
- Design and development of the approach
- Implementation of a prototype
Future work
- Development of a production-ready installation package
- Optimized GPU acceleration of core functionality
- Validation against experimental data
Related Publication
Zaytsev, Y. V, Morrison, A., & Deger, M. (2015). Reconstruction of recurrent synaptic connectivity of thousands of neurons from simulated spiking activity. Journal of Computational Neuroscience, 39(1), 77–103. http://doi.org/10.1007/s10827-015-0565-5
Simlab Contact
Project Partner
Moritz Deger - University of Cologne