Robust decoding in a network of retinal ganglion cells sharing
common input
In preparation; presented as talks at COSYNE and SAND meetings
Synchronized firing among retinal ganglion cells (RGCs) has been
reported in many studies. Two major candidate mechanisms of
synchronized firing are direct coupling between the cells and common
input to the cells. Recent experimental work (Khuc-Trong and Rieke,
2008) indicates that electrical coupling between parasol cells is
weak, and neighboring parasol cells share significant synaptic input
even in the absence of modulated light stimuli. These findings suggest
that an accurate model of synchronized firing must include the effects
of common input. Here, we develop a new model of synchronized firing
that incorporates the effects of common input, and use it to simulate
the light responses and synchronized firing of a nearly complete
network of 279 simultaneously recorded parasol cells. We use a
generalized linear model augmented with a state-space model to infer
common input, spatio-temporal light response properties, and
post-spike history effects. The model captures the statistical
structure of the spike trains as well as the encoding of the visual
stimulus. We then proceed to decode the visual stimulus from the spike
train given the model parameters. We demonstrate that the common-input
model is much less sensitive to spike jitter than a model with direct
coupling.
Draft in progress (comments welcome) | Liam Paninski's home