SC98 IBM demo

Visualization of Action Potential Data.

Introduction
The plainfin midshipman (Porichthys notatus) is a sound-producing fish found along the intertidal regions of the Pacific Coast. Nest-building, male midshipman fish produce long duration (> 1 min) vocal signals known as hums to attract females to their nests. Hums are simple harmonic signals (2-3 harmonics) with a fundamental frequency near 100 Hz @ 16 degrees C. Males often congregate in a localized area and vocalize simultaneously, thus producing overlapping signals. Hence, in order to discriminate between the vocal signals of individuals and select a mate, females must first segregate the concurrent vocal signals within the chorus. The summation of two signals with slightly different frequencies produces a signal known as a beat. Beats are characterized by amplitude and phase modulations at a rate equal to the dif ference frequency (dF). Within a natural population of midshipman, differences in the fundamental frequencies of individual hums vary between 0-10 Hz at a given temperature (Bodnar and Bass 1997).

Neurons within the auditory midbrain synchronize (phase lock) their spike outputs to beats with dFs between 2-10 Hz. Some units appear to be tuned to specific dFs (Bodnar and Bass 1997). In a more recent study of the encoding of spectral differences in beat signals by midbrain neurons, we have found that while the spike rate responses of many neurons are sensitive to the spectral composition of a beat, virtually all midbrain units can encode information about differences in the spectral composition of beat stimuli via their interspike intervals (ISIs) with an equal distribution of ISI spectral sensitivity across the behaviorally relevant dFs (Bodnar and Bass, in press). Together, temporal encoding in the midbrain of dF information through spike synchronization and of spectral information through ISI could permit the segregation of concurrent vocal signals.

Further information may be found at a Bass lab site, and at a Cornell news site. Part of the news site has fish sounds.

Bodnar and Bass have found:

Neurons in the perhipery which encode the fundamentals by firing in phase with the 90 (+/-) Hz fundamentals. Neurons in the midbrain, however, do not phase-lock to the fundamentals.
Individual neurons in the midbrain produce action potentials which encode the phase of beat-frequencies by firing at certain phases.
Neurons in the midbrain which encode the fundamentals in the interspike-intervals. A given neuron may have a different distribution of ISIs for different combinations of fundamentals, which taken together with the beat-freqency phase-locking, gives enough information to specify both sources.

Data Analysis
Statistical reduction of the large number of action potential events and visualization of the data have been used as complimentary ways of understanding patterns in the dataset.

Details of the Vis
The data shown is for presentations of sound at 90 Hz plus either 96 Hz or 84 Hz, refered to as +6 and -6 respectively. The neurons studied were classified by:

Optimum frequency difference (df).
Individual cells responded best to different fundamental differences, although some responed to all differences. Since we are displaying data for +/-6 Hz, we would expect the 6 Hz optimum cells to respond the best.
Shape of the ISI distibution.
Individual cells showed ISI distributions which were different in the way they responded to 90 +/-6 Hz. Many cells showed a differential response over a specific range of ISIs. The cells were classified into four catagories:
- Short -- Cells which showed differential response in ISI at short ISI, typically around 5 msec.
- Meduim -- Cells with diffences around 15 msec.
- Long -- Cells with differences around 30 msec.
- Cells which showed no differential response.

The Y-axis corresponds to different optimum dfs. On the "slice display" the slice number shows all cells of the same optimum df. Slice number is:

0.
1.
2.
3.
4. Optimum df=6 Hz.
5.
6.
7.
8.

The X-axis is arranged to separate short, medium, long, and all ISI responses.

Animation

Combined data; time slice
Data combined from ten experimental runs is shown as a function of time. The size of the glyphs is proportional to the number of action potentials occuring during a 1 msec time bin. df axis is horizontal. Color of the glyphs is controlled by the phase of the stimulus.

Smoothed data; time slice
Data combined from ten experimental runs, averaged over 10 time steps, is shown as a function of time. The size of the glyphs is proportional to the number of action potentials occuring during a 10 msec time bin. df axis is horizontal. Color of the glyphs is controlled by the phase of the stimulus.

Smoothed data; 3D time progression
As seen above, it is hard to understand the occurance of action potentials by viewing a 'movie' progressing through time. In this animation, we grow a 3D volume representing (on 3 axes) df, ISI type, and time. The size of the glyphs is proportional to the number of action potentials occuring during a 10 msec time bin. Color of the glyphs is controlled by the phase of the stimulus.

Smoothed data; 3D rotation
The 3D representation that was 'grown' in the previous animation is rotated to show the structure. Note the strong phase-locking of all the cells to the 6 Hz beat-freqency. Total time of presentation is 1 second.

Model Cells; 3D Time Progression
A model of a ring of connected neurons passes action potentials around. Here we append time-slices to from a 3D data set.

Model Cells; 3D rotation
A ring of model neurons passes action potentials around a loop.

Model vs Real; 3D rotation
Data for +6 compared to model data.

Screen Shots of slices