Inferring synaptic inputs from spikes with a conductance-based neural encoding model
- University of Washington, United States
- Princeton University, United States
Figures
Figure 1 with 1 supplement
Model diagrams.
(A) Diagram illustrating novel biophysical interpretation of the generalized linear model (GLM). The stimulus is convolved with a conductance filter weighted by , the difference between …
Figure 1—figure supplement 1
Convergence of model parameter fits.
(A) Estimates (solid traces) of excitatory (blue) and inhibitory (red) filters from 10 min of simulated spike trains. (Dashed lines indicate true filters). The inhibitory filter was oppositely tuned …
Figure 2
Validating the LN conductance model.
The CBEM describes the relationship between stimulus and each synaptic conductance with a linear-nonlinear (LN) cascade, consisting of a linear filter followed by a fixed rectifying nonlinearity. (A)…
Figure 3
Validating the firing rate nonlinearity.
(A) Schematic of the mapping from membrane potential to spikes under the CBEM. (B) The raw (gray) and spike-triggered (black) distribution of intracellular membrane potential obtained from …
Figure 4
Predicting conductances from spikes with CBEM.
Model parameters and conductance predictions for two example ON parasol RGCs. Left: Linear kernels for the excitatory (blue) and inhibitory (red) conductances estimated from spike train data (light …
Figure 5
Summary of the CBEM fits to 6 ON parasol RGCs for which we had both spike train and conductance recordings.
(A) The correlation coefficient () between the mean observed excitatory synaptic input to a novel 6 s stimulus and the conductance predicted by the LN cascade fit to the excitatory conductance …
Figure 6 with 1 supplement
Summary of the CBEM fits to 5 ON midget RGCs.
The plot follows the same conventions as the parasol results in Figure 5. (A,B) The correlation coefficient () between the mean observed excitatory and inhibitory synaptic input to a novel 6 s …
Figure 6—figure supplement 1
CBEM fit for an example ON-midget cell with a comparison the LN models fit directly to the conductances.
Left: Linear kernels for the excitatory (blue) and inhibitory (red) inputs estimated from the conductance-based model (light red, light blue) and estimated by fitting a linear-nonlinear model …
Figure 7 with 1 supplement
CBEM spike train predictions.
(A) Spike rate prediction performance for the population of nine cells for 5 s test stimulus. The true rate (black) was estimated using 167 repeat trials. The red circle indicates the cell shown in …
Figure 7—figure supplement 1
Relation of inferred conductances to GLM prediction error.
(A) The difference in the GLM predicted rate () and the measure spike rate () from the PSTH in Figure 7c compared to the excitatory conductances predicted by the CBEM. We only considered the …
Figure 8
Comparison of CBEM and GLM fits.
(A) Spike rate prediction performance and (B) cross-validated log-likelihood for the population of nine cells for 7 s test stimulus for the GLM and the CBEM with only an excitatory input term (CBEMex…
Figure 9 with 2 supplements
Contrast gain control in the CBEM.
(A) GLM filters for an example ON cell fit to responses recorded at 24%, 48%, and 96% contrast. (B) GLM filters fit to spike trains simulated from the CBEM fit to the cell shown in A. The CBEM was …
Figure 9—figure supplement 1
The filter heights (the absolute value of the peak of the filter) of the GLM fits to eight cells at all three contrast levels (one point per contrast level per cell; lines connect all contrast points from a cell), compared to the GLM filters fit the CBEM simulations of those same cells.
Figure 9—figure supplement 2
Correlation between the CBEM's excitation and inhibition depends on contrast.
(A) The average CBEM cross-correlation between excitation and inhibition for the 5 OFF cells fit to multi-contrast stimuli. The cross-correlations were computed from 12% (lightest trace) to 100% …
Figure 10
CBEM fits to a population of 27 RGCs.
(A) Temporal profile of the excitatory (blue) and inhibitory (red) at the center pixel of the receptive field for 16 OFF parasol cells. The thick lines show the mean. (B) The mean spatial profiles …
Figure 11
Predicted responses to spatially structured stimuli.
(A) Example sequences of 5 × 5 pixel frames of three different types of spatiotemporal noise stimuli used to probe the CBEM. The spatio-temporal stimulus was the same binary noise stimulus used to …
