https://hjjimmykim.github.io/HeadDirectionCells/
This project is an interactive visual demonstration of the data collection process for my research on neural decoding. A major function of the nervous system involves representing external stimulus through the collective activity of a population of neurons. This process, known as neural encoding, can be characterized by what is called a tuning curve, which gives each neuron's expected response (firing rate) as a function of the stimulus. Usually, a tuning curve takes the form of a unimodal curve (often approximated by a Gaussian) centered around the neuron's preferred stimulus value. In order for a neural code to be useful, it must be possible to perform the inverse process of neural decoding - inferring the original stimulus from the neural response. The mean deviation between the true stimulus and the decoded stimulus can be taken as a measure for the effectiveness of a given encoding strategy. My interest is in determining how this expected error varies with tuning curve parameters, in particular its width.
This is a fairly abstract process involving only the numerical simulations and statistical inference. Therefore, in order to facilitate visualization, I decided to choose a tangible real-life example. Head direction cells, as the name suggests, encode the direction (in a 2D plane) towards which the head is pointing. As each neuron has a preferred direction for which it is most responsive, a sensible approach was to have them organized in a ring. The color of the neurons was then used to represent their expected amount of activity, determined by both the input direction as well as the tuning curve parameter. With the view of allowing the user to gain a sense of this relationship, both factors have been set to be adjustable. Finally, the user can also run a graphical simulation of sampling actual neural responses with Poisson statistics, which are then decoded with a simple population vector method (where the decoded direction is given by the vector sum of the preferred directions, weighed by the corresponding neuronal activity). The resulting mean squared error, once it converges with enough number of trials, constitutes a data point for the actual research.
The project was written entirely in HTML, CSS, and JavaScript supplemented with the D3 library. It is essentially self-contained and requires no external data. Some bugs and clunkiness are to be expected in the light of my lack of prior experience with the tools used. I hope to further improve and streamline the product as I believe this could be a valuable visual aid for describing my research as well as the basic concept of neural coding.