Neuronify: An Educational Simulator for Neural Circuits

Abstract

Educational software (apps) can improve science education by providing an interactive way of learning about complicated topics that are hard to explain with text and static illustrations. However, few educational apps are available for simulation of neural networks. Here, we describe an educational app, Neuronify, allowing the user to easily create and explore neural networks in a plug-and-play simulation environment. The user can pick network elements with adjustable parameters from a menu, i.e., synaptically connected neurons modelled as integrate-and-fire neurons and various stimulators (current sources, spike generators, visual, and touch) and recording devices (voltmeter, spike detector, and loudspeaker). We aim to provide a low entry point to simulation-based neuroscience by allowing students with no programming experience to create and simulate neural networks. To facilitate the use of Neuronify in teaching, a set of premade common network motifs is provided, performing functions such as input summation, gain control by inhibition, and detection of direction of stimulus movement. Neuronify is developed in C++ and QML using the cross-platform application framework Qt and runs on smart phones (Android, iOS) and tablet computers as well personal computers (Windows, Mac, Linux).

Keywords: app; modeling; neural networks; software; teaching.

Figure 1.

Step-by-step illustration of how to build a simple neural circuit in Neuronify. A, A neuron is added to the canvas by dragging it from the creation menu. B, A DC current source is added and connected to the neuron by dragging the DC current source connection handle onto the neuron. C, A voltmeter is added and connected to the neuron by dragging the voltmeter connection handle onto the neuron. D, The properties of neurons and other items can be changed in the properties panel.

Figure 2.

Neuronify workspace. Here, a simulation has been loaded where two touch input sensors are connected to one excitatory neuron (A) and one inhibitory neuron (B). Neuron C is connected to a voltmeter that plots the membrane potential as described by the integrate-and-fire model. This network can be used to illustrate how neuron B can inhibit neuron C so that when neuron A fires shortly after, A may not be able to excite neuron C beyond its threshold potential. Activating neuron A results in a spike in neuron C (first spike in the figure). However, if neuron B is activated first and then neuron A shortly after, neuron C is not excited beyond its threshold potential. To the right we see the toolbar that overlays the workspace and at the bottom we see the playback controls.

Figure 3.

Menus in Neuronify. A, Main menu. B, Creation menu. C, Playback controls. D, Properties panel.

Figure 4.

A, Leaky integrate-and-fire neuron. The membrane potential of a leaky neuron is shown as plotted by the voltmeter item. As can be seen, the membrane potential increases until it reaches its threshold value and is immediately reset to the initial potential. The spike itself is overlayed as a vertical line for illustrative purposes and is not explicitly included in the dynamics of the membrane potential. B, Adaptive leaky integrate-and-fire neuron. The membrane potential of an adaptive neuron as plotted by the voltmeter item. This neuron receives input from the same DC current source. The interval between each spike of the adaptive neuron increases due to the an additional hyperpolarizing current, which grows for each spike and decays between the spikes.

Figure 5.

Illustration of receptive fields implemented in the visual input item in Neuronify. The user may choose between these and use them in combination with input from the camera on their device to simulate a neuron with a visual receptive field. A, Rectangular edge-detecting receptive field. B, Circular center-surround receptive field. C, Orientation-selective receptive field.

Figure 6.

Example of how Neuronify can be used to create interactive illustrations for neuroscience courses. This is a reproduction of figure 8.5 in Sterratt et al., 2011. The example shows how different levels of current injection into a neuron model results in different firing rates. Note that this example uses an artificial resting potential of 0 mV.

Figure 7.

Example illustrating integration of synaptic inputs. In the upper circuit, the output neuron only receives input from a single presynaptic neuron. This input alone is not sufficient to make the output neuron spike. In the lower circuit, the output neuron instead receives input from three presynaptic neurons. This makes the neuron fire, thus illustrating how a neuron effectively integrates the synaptic input it receives to produce spikes. In the app, this example uses touch sensors instead of a current source for a more interactive illustration of this behavior.

Figure 8.

Example of gain control with feedback inhibition. The input neuron receives a constant direct current input and is connected to neuron A, which in turn is connected to the output neuron. The output neuron is further connected to the inhibitory neuron B. Neuron B inhibits neuron A, which in total results in feedback inhibition, i.e., reduced activity in the output neuron compared with the input neuron.

Figure 9.

Example of direction-selective network. This example illustrates a direction-selective feedforward network based on one-sided lateral inhibitory connections. The upper row of touch inputs are connected to the input neurons. These are both connected to the relay neurons and the inhibitory neurons. Each inhibitory neuron inhibit the relay neuron positioned immediately to the right in the network. The relay neurons are connected to the output neuron. The effect of the inhibition is that the network only responds to input where the touch sensors are pressed sequentially from right to left but not in the opposite direction.