Create a data set. Do this by creating a standard (number or letter) code for the neuronal responses you have gathered. For example, if the neural code information you have gathered is based on the number of times the neuron fires within one minute of the stimulus, then you could just use the number (of pulses per minute) as the code. So, your data set may list stimulus type and rate of firing as such:
A - 16
B - 2
C - 5
AB - 10
AC - 20
BC - 3
Include multiple results for each stimulus type.
Find a neural network model to use. As you are attempting to predict neural output (the neural codes), be sure to use biological software that takes into account the properties of the neural system.
Use backpropagation to train the neural network. Split your data set into two smaller sets. Use about 90 percent of your data set as a training set. Save the remaining data for later to use as a validation data set. Enter your training data set into the neural network. Using backpropagation, the neural network can quickly test its predictions against the actual data many times over, each time coming closer to an effective neural predictor.
Enter your validation data set into the trained network. Once the neural network has gone through the training data set and found a neural predictor, use the remaining data you have to test the predictor and ensure it works well. Validation will check to see that the predictions made by the neural predictor match the responses you actually found in your experiments.
Enter new hypothetical data with unknown responses into the trained and validated neural network. The responses you receive are the predicted results of the hypothetical experiment.