Neural coding 1: How to understand what a neuron is saying.

In this post I am diverting from my usual astrophotography theme and entering the world of computational neuroscience, a subject I studied for almost ten years. Computational neuroscience is a relatively new interdisciplinary branch of neuroscience that studies how areas of the brain and nervous system process and transmit information. An important and still unsolved question in computational neuroscience is how do neurons transmit information between themselves. This is known as the problem of neural coding and by solving this problem, we could potentially understand how all our cognitive functions are underpinned by neurons communicating with each other. So for the rest of this post I will attempt to discuss how we can read the neural code and why the code is so difficult to crack.

Since the twenties we have known that excited neurons communicate through electrical pulses called action potentials or spikes (see Figure 1). These spikes can quickly travel down the long axons of neurons to distant destinations before crossing a synapse and activating another neuron (form more information on neurons and synapses see here).

Figure 1. Neural action potentials. An action potential diagram is shown on the left as if recorded from inside a neuron (see inset). For an action potential arise and propagate through a neuron, it must reach a certain threshold (red dashed line). If it doesn’t the neuron will remain at rest. The right panel shows a real neurons firing spikes in the cortex of a mouse. Taken from Gentet LJ et al. (2010).

You would be forgiven for thinking that the neural coding problem is solved: neurons fire a spike when the see a stimulus they like and communicate this fact to other nearby neurons, while at other times they stay silent. Unfortunately, the situation is a bit more complex. Spikes are the basic symbol used by neurons to communicate, much like letters are the basic symbols of a written language. But letters only become meaningful when many are used together. This analogy is also true for neurons. When a neuron becomes excited it produces a sequence of spikes that, in theory, represent the stimuli the neuron is responding to. So if you can correctly interpret the meaning of spike sequences you could understand what a neuron is saying. In Figure 2, I show a hypothetical example of a neuron responding to a stimulus.

Figure 2. A stimulus (top trace) fluctuates over time (s(t)) and spikes from a hypothetical neuron are recorded. The stimulus is repeated 5 times producing 5 responses r1,2,3…5 shown below the stimulus. Each response is composed of spikes (vertical lines) and periods of silence. By counting the number of spikes within small time window lasting Δt seconds, we can calculate the firing rate of the neuron (bottom trace).

In this example a neuron is receiving an constantly fluctuating input. This is a bit like the signal you would expect to see from a neuron receiving a constant stream of inputs from thousands of other neurons. In response to this stimulus the receiving neuron constantly changes its spike firing rate. If we read this rate we can get a rough idea of what this neuron is excited by. In this case, the neuron fires faster when the stimulus is high and is almost silent when the stimulus is low. There is a mathematical method that can extract the stimulus that produces spikes, known as reverse correlation (Figure 3).

Figure 3. Reverse correlation can identify what feature of the stimulus (top) makes a neuron fire a spike (bottom). Stimulus samples are taken before each spike (vertical lines) and then averaged to produce a single stimulus trace representing the average stimulus that precedes a spike.

The method is actually very simple; each time a spike occurs we take a sample of the stimulus just before the spike. Hopefully many spikes are fired and we end up with many stimulus samples, in Figure 3 the samples are shown as dashed boxes over the stimulus. We then take these stimulus samples and average them together. If these spikes are being fired in response to a common feature in the stimulus we will be able to see this. This is therefore simple method of finding what a neuron actually responds to when it fires a spike. However, there are limitations to this procedure. For instance, if a neuron responds to multiple features within a stimulus then these will be averaged together leading to a misleading result. Also, this method assumes that the stimulus contains a wide selection of different stimulus fluctuations. If it doesn’t then you can never really know what a neuron is really responding to because you may not have stimulated it with anything it likes!

In my next two posts, I will discuss how more advanced methods from the realms of artificial intelligence and probability theory have helped neuroscientists more accurately extract the meaning of neural activity.

Post by: Daniel Elijah

2 thoughts on “Neural coding 1: How to understand what a neuron is saying.”

  1. Is it possible that two kinds of information are being processed simultaneously by the brain? Most of capacity is dedicated to homeostasis and housekeeping. Usually this means a signal that communicates a function mechanistically from one cell to another. However in the case of senses, a stimulated cell mechanistically communicates a bit of factual information. Each photo-receptor signal communicates the presence of a specific light frequency, i.e.: 450Hz. After being processed by parallel functions in sight-facilitating regions, the signal stimulated by 450 Hz light, along with millions of other photo-receptor signals, present altogether as a photo impression symbol of visual reality in consciousness. I suggest that cognitive brain regions work with enormous mega-signal impression symbols as a whole. This would be somewhat like a megapixel digital photo is processed in a camera. In such a symbol, the 450 Hz signal acts as a red pixel that may be part of a spot of blood in the context of a larger scene. Signals forming large impression symbols of reality and signals orchestrated to cause body functions have different kinds of information that may confuse scientists trying to discover what is occurring.

  2. So far so good, but a couple of important things are left out. One is that the information (code or whatever you want to call it) is represented in circuits not just individual neurons. Secondly, inter-spike intervals contain information (even in the Shannon sense). See my recent book, “Mental Biology.”

Comments are closed.

Share This