This is the cover of a CD I made recently which takes neural ensemble recordings of the moth antennal lobe as it responds to odor and converts it to music.   Tracks: Factor 1: 5 neurons with factor Factor 2: 4 neurons with factor Factor 3: 4 neurons alone Factor 4: 9 neurons with factor 3/4 real time: 2-hexanone 3/4 real time: 2-heptanone
Back Cover	Back to Kevin's home

A model system for the study of olfactory processing
The image of Leonardo Di Vinci’s Vitruvian Man is believed to be an artistic expression of an algorithm created by Di Vinci to solve the problem known as “squaring the circle.” Squaring the circle is an ancient geometrical problem whereby of a pair of compasses and a ruler are used in an attempt to construct a circle and square of equal area. In many respects the comparative approach to understanding the function of the human brain is analogous. In our case we are “manning the moth.” That is, we study the structure and function of the moth brain in order to better understand our own. The Sphinx moth, Manduca sexta is well known for its highly sensitive sense of smell. The antennal lobe (featured on the cover) is the part of the moths nervous system that processes odor. Of particular importance is the amazing degree of similarity of this antennal lobe's circuitry to our own odor processing center, the olfactory lobe. This similarity provides us with opportunity for intensive study of the physiology of odor processing.

About the soundtracks
The tracks contained on this CD was created from a single recording of 32 simultaneously monitored neurons and their responses to 20 100ms presentations of 9 different but highly related odors. The odors, which are played back in the following order, 1-hexanol, 1-heptanol, 1-octanol, 1-nonanol, 1-decanol, 2-hexanone, 2-heptanone, 2-octanone and 2-decanone. vary by single carbon units (CH3) or they contain different functional groups. Odors ending in anol are alcohols and those ending in anon are ketones). For each of the 180 100ms odor presentations a 780ms recording was made for each neuron starting at the moment the odor was released. This 780ms response window was then sliced into 32 equal 30ms pieces of time and the number of nerve impulses, or action potentials, was counted for each slice. These counts typically sum to zero because 30ms (3/100 seconds) is fairly brief. However, because the neuron can burst the counts occasionally reach a number as high as 6 or 7. Thus, the 780ms response window is converted into a sequential string of 26 numbers, each number representing one slice, and each number ranging from 0 to ~7. Before the data was converted to sound principle components used to sort neurons into groups based on synchronous activity across the 26 slices and across all odors. This analysis produces a single factor with a new string of 26 numbers that represents correlated activity of the neurons in that group. The original data and these factors are then imported into a program (http://algoartcom) that converts the range of numbers in the string into MIDI range (0-127).

Next, the software allows the programmer to assign an instrument, an octave range and a volume to each neuron and/or factor. The programmer may also use one of a number of functions to convert the raw MIDI data into something more patterned. However, for the tracks on this CD only the quantize function was used, allowing neurons to be “played” in major, minor etc; nothing more was done. Thus, each resulting track is a fairly straight forward conversion of raw neurophysiological data into sound. Each track is a subset of the 32 recorded neurons that were identified by the principle components analysis as a group of synchronously active neurons. In the first 3 tracks, the factor was included and provides a melody line where as the neurons are represented by unique instruments. The first 4 tracks are the sequential presentations of the 20 presentations of each odorant played at about 1/6th real time. The last 2 tracks are continuous recordings of 10 puffs of the same odor each signified by a bong sound and separated by about 4sec (1/2 real time); these are the responses of all 32 neurons, each neuron is represented by a single tone.

Why do neurones sound like music?
The short answer is; time matters. I assure you I know almost nothing about music. However I do at least know that there are some basic characteristics common to most music; pattern repetition and evolution. The repetition in these songs comes from repeated presentations of the same odor; the evolution comes from changing the features of the odorant being presented. If odorant did not produce a consistent patterned response across the 780ms there would be little repetition over the 20 consecutive presentations of same odorant. If the neurons did not respond with a different pattern for at least some of the odorants then there would be no evolution. Thus, each recording at least approximates music because there is a consistent patterned output from each of the individual neurons in each track. Keep in mind that in the brain all of these songs were playing simultaneously.

Acknowledgments:
Nicolas Gibson produced the cover image; the antennal lobe of Manduca sexta
Leonardo Di Vinci the Vitruvian Man; sorry for the moth graffiti.
Thanks to professors John G. Hildebrand, Tom A. Christensen and Brian H. Smith.
This work was supported by grants from NIH/NIDCD to KCD and BHS.

Back to Kevin's home