Figure 1.
Schematic illustration of temporal partitioning schemes.
A) Consider a spike train within a time window T during stimulus presentation. The timing of individual spikes can be measured using a binning procedure relative to stimulus onset (gray boxes). Alternatively, the timing can be measured relative to an intrinsic slow oscillatory signal. Here the phase of such an oscillation was divided into four phase quadrants (φi) and spikes are color-coded by their respective phase angle. B) In this study we consider three codes. i) The spike count, defined as total number of spikes within the window T. ii) A code based on a spike train partitioned using stimulus-locked and equally-spaced time bins (‘time-partitioned’), defined as the vector consisting of the number of spikes per time bin. iii) A code based on the phase-partitioned spike train (‘phase-partitioned’), defined as the vector consisting of the number of spikes per phase range (e.g. phase quadrant, colored).
Figure 2.
Example data from an auditory cortex neuron illustrating responses in each code.
A) Sound wave of the 52 sec acoustic stimulation sequence consisting of various natural and environmental sounds. Dashed lines illustrate the random selection of 10 epochs from the long sound sequence as used for the decoding analysis (epoch duration not to scale). B) Data from one example neuron showing the spike raster (left) and the average response patterns for the three codes (right). The spike raster displays the response to multiple repeats of ten stimulus epochs (concatenated for display purposes). Spikes are color-coded with the phase of the theta (2–6 Hz) band oscillation at the time of spike. The average responses illustrate the trial-averaged responses for each code. Colors indicate the phase bins, gray-scales indicate the time bins.
Figure 3.
Population results for auditory cortical data.
A) Decoding performance across neurons (n = 40, mean and s.e.m.) for each code (N = 8 bins, T = 160 ms window, 2–6 Hz LFP). B) Difference between the decoding performance in time- and phase-partitioned codes and the spike count for each neuron. Neurons are sorted according to the difference for the time-partitioned code (red). Parameters as in A). C) Histogram across neurons of the % gain in decoding performance in the dual time- and phase-partitioned code over the better (for each neuron) of the two individual codes. Parameters as in A). D) Correlation coefficient between the inter-trial phase coherence of the 2–6 Hz LFP and the decoding performance (percentage correct) for individual stimulus epochs used for decoding. Boxplots display the median and quartiles across neurons for each code. Parameters as in A). E) Population averaged decoding performance for different lengths of the stimulus epoch window T (N = 8 bins) and for different numbers of bins (T = 160 ms). F) Ratio of the information in the phase-partitioned code to the information in the spike count when using different frequency bands (4 Hz width) to derive the oscillation phase. Abscissa values indicate center frequencies for each band.
Figure 4.
Example and population data for visual cortex data.
A) Data from an example neuron showing the spike raster (left) and the average responses for the three codes (right). The spike raster displays the responses to multiple repeats of the video stimulus during ten selected epochs (concatenated for display purposes). Spikes are color coded with the phase of the theta (2–6 Hz) band LFP at the time of spike. The average responses illustrate the trial-averaged responses for each code. Colors indicate the phase bins, gray-scales indicate the time bins. B) Decoding performance across neurons (n = 37, mean and s.e.m.) for the three codes (N = 8 bins, T = 160 ms window, 2–6 Hz LFP). C) Difference between the decoding performance in time- and phase-partitioned codes and the spike count for each neuron. Neurons are sorted according to the difference for the time-partitioned code (red). Parameters as in B). D) Histogram across neurons of the % gain in decoding performance in the dual time- and phase-partitioned code over the better (for each neuron) of the two individual codes. Parameters as in B). E) Population averaged decoding performance for different lengths of the stimulus epoch window T (N = 8 bins) and for different numbers of bins (T = 160 ms). F) Ratio of the information in the phase-partitioned code to the information in the spike count when using different frequency bands (4 Hz width) to derive the oscillation phase. Abscissa values indicate center frequencies for each band.
Figure 5.
Response partitioning in face of temporal uncertainty.
A) Schematic of single trial epoch selection for the decoding process assuming a perfect temporal alignment across trials. When composing the codebook for decoding, the epochs for individual trials are all sampled at the same position relative to the stimulus presentation (blue). Hence the reference epoch in the codebook (blue) and the to-be-decoded single trial (black) are in perfect temporal alignment. B) Schematic for a decoding process introducing a temporal uncertainty (jitter) between trials when composing the codebook. The data epochs for individual trials were shifted by a lag value that was randomly sampled for each trial and which was uniformly distributed between −J/2 and +J/2, where J corresponds to the (maximal possible) temporal uncertainty. C) Decoding performance as a function of temporal uncertainty J for auditory cortical data (n = 40 units, T = 160 ms, N = 8, 2–6 Hz LFP). D) Decoding performance as a function of temporal uncertainty J for visual cortical data (n = 37 units, T = 160 ms, N = 8, 2–6 Hz LFP).
Figure 6.
Auditory coding in the presence of background noise.
A) Schematic of sound waveforms showing the target sound without background noise and with background noises of three levels. B) Left: Decoding performance across neurons (n = 43 units, mean and s.e.m.) for the three different codes as a function of noise level. Right: Decoding performance with the spike count and the phase-partitioned code for individual neurons (N = 8 bins, T = 160 ms, 2–6 Hz LFP).