Model Tau Spread with Brain Connectome Data

ISEF Category: Cellular and Molecular Biology

Subcategory: Neurobiology  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Alzheimer's does not spread evenly through the brain. It moves along circuits, almost like a rumor traveling through a social network. That makes it possible to model the disease with graphs, data, and careful validation. If you like coding and biology, this topic gives you both.

What Is It?

Tau is a protein that can misfold and build up inside neurons. In Alzheimer's disease, tau pathology tends to appear in a staged pattern, often described by Braak stages. Scientists use that pattern to study how damage moves through the brain over time.

A connectome is a map of how brain regions connect to each other. The Allen Brain Atlas gives researchers rich brain data, including gene expression and cell-type context for some analyses. In this project, you treat the brain like a network graph, where regions are nodes and connections are edges. Then you test whether a plain connectivity-based model predicts tau spread well, or whether adding microglial-density priors gives you better predictions. Microglia are the brain's immune cells, and they may shape how pathology grows or clears.

Think of it like predicting how a stain spreads through a sponge. The sponge's structure matters, but so does what the sponge is made of. Your question asks whether structure alone explains the pattern, or whether cell biology improves the forecast.

Why This Is a Good Topic

This makes a strong science fair topic because you can build a clear model, compare two competing hypotheses, and measure which one performs better. The project connects to a real biomedical problem, Alzheimer's progression, and to a real research skill, computational prediction on biological networks. You can learn network analysis, model validation, and statistical comparison without needing wet lab work.

Research Questions

• How does a connectome-only model predict Braak-stage progression across brain regions??
• What is the effect of adding microglial-density priors on model accuracy for tau-spread prediction??
• Does a weighted network model outperform an unweighted network model in predicting regional tau burden??
• To what extent do different connectivity measures change the fit between predicted and observed Braak stages??
• Which brain regions are misclassified most often by a connectome-only model??
• How does model performance change when you test it on held-out regions or held-out subjects??

Basic Materials

• Laptop with enough memory to run Python or R.
• Allen Brain Atlas data access through a browser or downloadable files.
• Public tau staging or Alzheimer's progression dataset from a peer-reviewed source.
• Spreadsheet software for cleaning metadata and tracking samples.
• Citation manager for organizing papers.
• External hard drive or cloud storage for versioned backups.

Advanced Materials

• University workstation or lab computer with stronger RAM and CPU.
• Access to neuroimaging or transcriptomic datasets linked to Braak staging.
• Brain connectivity matrices from the Allen Brain Atlas or related open datasets.
• Region-level microglial density or proxy datasets from published studies.
• Version-control system for code and analysis logs.
• Statistical computing environment for network modeling and validation.

Software & Tools

• Python: Builds network models, runs prediction code, and handles data cleaning.
• R: Tests model differences and makes publication-style plots.
• Jupyter Notebook: Keeps code, notes, and figures in one reproducible file.
• NetworkX: Represents the brain as a graph and computes network features.
• ImageJ: Helps if you need to inspect or quantify figures from published brain maps.

Experiment Steps

1. Define the prediction target, such as Braak stage, regional tau burden, or stage ordering across regions.
2. Choose your brain network representation, including which regions count as nodes and which connection metric counts as an edge.
3. Build a baseline model that uses connectome structure alone, so you have a clean comparison point.
4. Add microglial-density priors as a second model input and decide how you will encode them mathematically.
5. Plan a validation scheme that tests prediction on held-out regions, subjects, or atlas partitions.
6. Predefine the metrics you will compare, such as accuracy, rank correlation, AUC, or error by stage.

Common Pitfalls

• Using a connectivity map with a different region naming scheme than the tau dataset, which breaks the alignment between inputs and outcomes.
• Mixing human and mouse sources without checking whether the regions and biology match, which makes the comparison hard to defend.
• Treating microglial density as a direct measurement when the source paper only provides a proxy or an estimate.
• Fitting and testing on the same regions, which makes the model look better than it really is.
• Ignoring class imbalance across Braak stages, which can hide weak performance on the rare stages.

What Makes This Competitive

A class-level version of this project stops at building one model. A stronger version compares multiple network formulations, tests whether microglial priors add value beyond simple connectivity, and uses out-of-sample validation. You can push it further by checking whether the result holds across different atlases, different staging schemes, or different performance metrics. Clear uncertainty estimates and thoughtful controls matter more than flashy code.

Project Variations

• Test whether adding astrocyte-related priors changes prediction as much as microglial priors do.
• Compare human connectome-based predictions with mouse connectome-based predictions to see which better matches Braak-like staging.
• Replace stage prediction with regional ordering, and test whether the model can recover the sequence of tau spread.

Learn More

• Allen Brain Atlas: Search the atlas site for brain connectivity, gene expression, and cell-type resources tied to region-level brain analysis.
• PubMed: Search for review articles on tau propagation, Braak staging, and microglia in Alzheimer's disease.
• NIH NCBI Bookshelf: Find free chapter-length background on neurodegeneration, protein aggregation, and brain immune cells.
• MIT OpenCourseWare: Look for free courses on computational biology, graph theory, or machine learning that help with model building.
• Nature Reviews Neuroscience: Search the journal for review articles on Alzheimer's spread, connectomics, and neuroinflammation.