1. TAC Meeting Neuronal Coding in the Retina and Fixational Eye Movements Neuronal Coding in the Retina and Fixational Eye Movements 16.07.2009 Christian Mendl, Tim Gollisch Lab

2. Outline Experimental Setup Fixational Eye Movements Research Questions A look at the observed data Information theory: entropy, mutual information, synergy,... Outlook

3. Experimental Setup The retina is a complex cell network consisting of several layers: rods/cones, horizontal cells, bipolar cells, amacrine cells, and retinal ganglion cells input-output relationship? Multi-Electrode Array spikesorting

4. Fixational Eye Movements Retinal eye movement amplitudes approximately 5µm, corresponds to diameter of a photoreceptor Eye movements of the turtle during fixation Greschner M, Bongard M, Rujan P, and Ammermüller J. Retinal ganglion cell synchronization by fixational eye movements improves feature estimation. Nature Neuroscience (2002) source: Martinez-Conde laboratory

5. Research Questions Main line of investigation: Image feature discrimination and fixational eye movements Concrete task: based on the spike responses from retinal ganglion cells, discriminate 5 different angles of a black-white border presented to the retina Wobbling border imitates fixational eye movements Optimal decoding strategy for stimulus discrimination? Role of population code? Green ellipses denote the receptive fields of 2 ganglion cells; blue arrow shows the wobbling direction

6. Observed Data stimulus period: 800 ms amplitude: 100µm, angle: 0.2·2πamplitude: 100µm, angle: 0.8·2π each dot represents a spike Spike timing correlations can provide information about the stimulus

7. Spike Timing Correlations amplitude: 100 µm, binsize: 50 ms, stimulus period: 800 ms shuffled correlations look similar, intrinsic interactions don‘t seem to be important receptive field centers and wobbling border angles histogram plot of relative spike timings

8. Binning the Spike Train stimulus-locked binning unlocked binning encoding the spike pattern → for either 0, 1 or 2 spikes in one bin, this results in 3 8 different patterns the pattern window is shifted by the stimulus period → observer knows the stimulus phase

9. Applying Information Theory Elad Schneidman, William Bialek, and Michael J. Berry. Synergy, Redundancy, and Independence in Population Codes. The Journal of Neuroscience (2003) Mutual information: Synergy: Quantify population responses by information theory measures (can be positive or negative)

10. Entropy Bias Correction Choose a close to optimal prior in Bayesian probability calculus to estimate the entropy of discrete distributions yields an entropy variance estimate IIlya Nemenman, Fariel Shafee, and William Bialek. Entropy and Inference, Revisited. In T. G. Dietterich, S. Becker, and Z. Ghahramani, editors, Advances in Neural Information Processing Systems 14, Cambridge, MA (2002). MIT Press. Main idea: extrapolate entropy to inverse data fraction zero Can be combined with NSB entropy estimation Strong, S. P.; Koberle, R.; de Ruyter van Steveninck, R. R. & Bialek, W. Entropy and Information in Neural Spike Trains Physical Review Letters, 1998, 80, 197-200 Probability distribution p exp estimated from finite data may omit rare events → corresponting entropy S(p exp ) is typically higher than the true entropy

11. Mutual Information (Individual Cells) unlocked binningstimulus-locked binning theoretical upper bound statistics for several cells

12. Mutual Information (Cell Pairs) individual cells

13. Quantifying the Population Code: Synergy redundancy Synergy versus mutual information for several recordings

14. Outlook Increase discrimination difficulty: – smaller or more angles – lower light intensity – grating instead of fixed border Effect of shorter stimulus periods and smaller wobbling amplitudes? Try different decoding stategies Neuronal network statistics – pairwise interactions sufficient to capture population statistics? Future projects: – try to capture observed data by neuronal models – biological counterparts? Elad Schneidman, Susanne Still, Michael J. Berry and William Bialek. Network Information and Connected Correlations. Physical Review Letters (2003)

15. Observed Data stimulus period: 800 ms amplitude: 100µm, angle: 0.2·2πamplitude: 100µm, angle: 0.8·2π

16. Observed Data (cont.) stimulus period: 800 ms amplitude: 100µm, angle: 0.4·2π amplitude: 100µm, angle: 0.6·2π

17. Observed Data (cont.) stimulus period: 800 ms amplitude: 100µm, angle: 0

18. Intrinsic Interactions ΔI signal versus ΔI noise. The former measures the effect of signal-induced correlations on the encoded information, whereas the later quantifies the contribution of intrinsic neuronal interactions to the encoded information.

19. Ising Model and Marginal Distributions Elad Schneidman, Susanne Still, Michael J. Berry and William Bialek. Network Information and Connected Correlations. Physical Review Letters (2003) Elad Schneidman, Michael J. Berry II, Ronen Segev and William Bialek. Weak pairwise correlations imply strongly correlated network states in a neural population. Nature (2006) Jonathon Shlens, Greg D. Field, Jeffrey L. Gauthier, Matthew I. Grivich, Dumitru Petrusca, Alexander Sher, Alan M. Litke, and E. J. Chichilnisky. The Structure of Multi-Neuron Firing Patterns in Primate Retina. Journal of Neuroscience (2006)

20. Preliminary Results: Connected Information Linear Ramps, frog recording

21. Preliminary Results: Connected Information (cont.) > 10% connected information of order 3 Linear Ramps, p. Axolotl recording

22. Ising Model and Marginal Distributions (cont.) In the perturbative regime, Δ N increases linearly with N and thus does not provide much information about the large N behavior Roudi Y, Nirenberg S, Latham PE (2009) Pairwise Maximum Entropy Models for Studying Large Biological Systems: When They Can Work and When They Can’t. PLoS Comput Biol 5(5): e1000380.

23. Preliminary Results: Perturbative Regime of Pairwise Models Roudi Y, Nirenberg S, Latham PE (2009) Pairwise Maximum Entropy Models for Studying Large Biological Systems: When They Can Work and When They Can’t. PLoS Comput Biol 5(5): e1000380.

24. Simple LN-Model

25. Preliminary Results: Spiking Latency need 3 cells to reconstruct 5 angles Elad Schneidman, William Bialek, and Michael J. Berry. Synergy, Redundancy, and Independence in Population Codes. Journal of Neuroscience (2003) Tim Gollisch, Markus Meister. Rapid neural coding in the retina with relative spike latencies. Science (2008)