Tinnitus Spiking Network Model

ISEF Category: Computational Biology and Bioinformatics

Subcategory: Computational Neuroscience  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A damaged ear can still hear sound that is not there. That phantom tone is tinnitus, and researchers think network activity in the brain may help drive it. You can model that idea with a spiking neural network and test whether stronger inhibition quiets the phantom signal. This project gives you a real neuroscience question and a clean way to measure it.

What Is It?

A spiking neural network is a computer model made of neurons that fire brief electrical pulses, or spikes. Instead of pretending the brain is one smooth signal, it tracks who fires, when they fire, and how strongly they influence each other. Brian2 is a Python tool that helps you build these models without writing every low-level detail from scratch.

For this topic, you simulate what happens after cochlear hair-cell loss. Hair cells in the inner ear turn sound vibrations into nerve signals. When some of them are gone, the brain may compensate in ways that increase spontaneous activity, which could help create a phantom tone. Think of it like a band where some instruments drop out, then the remaining players get turned up too far and start making noise on their own.

You then test a control idea, targeted lateral inhibition. Lateral inhibition means active neurons suppress nearby neurons, which can sharpen signals and reduce background noise. Your job is to see whether changing that inhibition lowers the model’s phantom-tone activity, and under what conditions the effect works best.

Why This Is a Good Topic

This is a strong science fair topic because you can change one variable at a time, measure a clear output, and compare several model designs. It connects to a real health problem, tinnitus, that affects millions of people. You can learn neural modeling, parameter sweeps, spike analysis, and data visualization without needing a wet lab. The project also leaves room for original work, since you can test different network structures, inhibition strengths, and loss patterns.

Research Questions

• How does increasing lateral inhibition change the rate of spontaneous phantom-tone spiking after simulated hair-cell loss?
• What is the effect of different degrees of cochlear hair-cell loss on network-wide baseline firing patterns?
• Does targeted inhibition around hyperactive neurons reduce synchronized firing more than uniform inhibition?
• To what extent do changes in excitatory-to-inhibitory balance alter the strength of phantom-tone activity?
• Which hair-cell loss patterns produce the largest increase in spontaneous rhythmic activity in the model?
• How does network noise level affect whether lateral inhibition can suppress tinnitus-like spiking?

Basic Materials

• Laptop or desktop computer with enough memory to run Brian2 simulations.
• Python installed with Brian2, NumPy, SciPy, and Matplotlib.
• Code editor such as VS Code, Spyder, or Jupyter Notebook.
• Spreadsheet software for organizing simulation parameters and outputs.
• Notebook for tracking model assumptions, parameter choices, and run labels.

Advanced Materials

• Access to a workstation or university computer cluster for larger parameter sweeps.
• Python with Brian2, Pandas, Seaborn, and statsmodels.
• Version control software such as Git for tracking code changes.
• R or Python stats packages for permutation tests or mixed-effects modeling.
• Optional plotting tools for raster plots, spike histograms, and heat maps.

Software & Tools

• Brian2: Builds and runs spiking neural network models for different neuron and synapse settings.
• Python: Handles simulation logic, parameter sweeps, and data analysis.
• Jupyter Notebook: Keeps code, notes, and plots in one place while you test model ideas.
• Matplotlib: Creates raster plots, firing-rate graphs, and comparison figures.
• Pandas: Organizes simulation outputs into tables for statistical analysis.

Experiment Steps

1. Define the tinnitus proxy you will measure, such as spontaneous firing rate, synchrony, or rhythmic burst activity.
2. Choose the network structure you will compare, including which neurons receive lateral inhibition and which do not.
3. Decide the hair-cell loss conditions you will simulate, so you can compare mild, moderate, and severe input loss.
4. Build a baseline model and check that it produces stable activity before you add damage or control logic.
5. Plan a parameter sweep for inhibition strength, then pair each run with the same noise and input settings for fair comparison.
6. Design the analysis pipeline before you run the full set, so you can compare spike rasters, summary rates, and statistical tests the same way every time.

Common Pitfalls

• Treating any increase in firing as tinnitus, which ignores whether the activity is coordinated enough to look like a phantom tone.
• Changing several network parameters at once, which makes it hard to tell whether inhibition or loss caused the effect.
• Using one random seed and calling it a result, which can hide how unstable the model is across runs.
• Forgetting to compare against a no-damage baseline, which makes it impossible to say what changed after simulated hair-cell loss.
• Picking a measurement that looks nice but does not match the model question, such as only averaging firing rate when synchrony is the real signal.

What Makes This Competitive

A strong version of this project goes beyond a single model run. You would compare several inhibition schemes, use repeated simulations, and show which network features predict phantom-tone activity most strongly. You could also test whether the same control works across multiple loss patterns, not just one. Clear statistics, thoughtful controls, and a model that explains when the effect fails would make the work much stronger.

Project Variations

• Test whether different patterns of hair-cell loss, like clustered versus scattered loss, change how easy it is to suppress phantom activity.
• Compare lateral inhibition with another control idea, such as changing excitatory gain or adding adaptive homeostatic scaling.
• Analyze whether the model’s results change when you measure synchrony, burst duration, or spectral power instead of only firing rate.

Learn More

• Brian2 Documentation: Free reference for building spiking neural network models in Python, found by searching the Brian2 docs site.
• Allen Brain Atlas: Data and resources on brain cell types and circuits, found on the Allen Institute website.
• NIH PubMed: Search for review articles on tinnitus, cochlear damage, and neural plasticity.
• MIT OpenCourseWare Computational Neuroscience: Free course materials that help you understand neural coding and network models.
• Journal of Neuroscience: Search review and research articles on tinnitus mechanisms and inhibition in auditory circuits.
• Python Scientific Computing Stack docs: Free documentation for NumPy, SciPy, Pandas, and Matplotlib, found through each project’s official site.