Early Alzheimer’s Multi-Omics Network Science Project

ISEF Category: Biochemistry

Subcategory: Other  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: Full Year

The Hook

One brain cell can leave a long fingerprint in its RNA and its metabolites. That makes early Alzheimer's look less like one broken switch and more like a miswired control panel. Public single-cell atlases let you study that fingerprint without a wet lab. You can turn public data into a disease map.

What Is It?

You can think of each cell like a tiny kitchen and control room. Transcriptomics tracks the RNA messages that tell the cell which recipes to run, while metabolomics tracks the small molecules that show what the cell is actually making, using, or storing.

When you put those two views together, you can look for co-regulation modules, groups of metabolites and genes that move together. In a healthy cell, those links may follow a stable pattern. In early Alzheimer's, some links may weaken, shift, or split apart, which can point to stress in specific cell types or brain regions.

Why This Is a Good Topic

This is a strong science fair topic because you can test a clear idea with public data, compare disease and control samples, and measure your results with statistics instead of guesswork. It also connects to a real medical problem, early Alzheimer's detection and cell stress, without needing a hospital or animal lab. You can learn data cleaning, normalization, network analysis, and how to judge whether a pattern stays real after you shuffle or resample the data.

Research Questions

• How does the strength of metabolite-gene co-regulation change between early Alzheimer's samples and matched controls?
• What is the effect of cell type on the number and density of metabolite-gene network modules?
• Does a novel correlation-network statistic separate disease samples from control samples better than a standard Pearson correlation network?
• To what extent do batch correction and normalization change the modules you detect in single-cell data?
• Which metabolites and genes stay linked across multiple donors, and which links disappear in early Alzheimer's?
• How does the network pattern differ across brain regions or tissue sources in public atlas data?

Basic Materials

• Laptop or desktop computer with at least 8 GB RAM.
• Reliable internet access for downloading public datasets.
• Python 3 with Jupyter Notebook installed.
• Spreadsheet software such as Google Sheets or Excel for tracking metadata.
• Free cloud storage or an external drive for backups.
• Headphones or a quiet workspace for long coding sessions.

Advanced Materials

• High-memory workstation with 32 GB RAM or more.
• Python environment with Scanpy, pandas, NumPy, SciPy, and NetworkX.
• R environment with Bioconductor packages for single-cell analysis.
• Institutional Linux server or cluster access for large matrices.
• External SSD for cached datasets and intermediate files.
• Secure research storage for donor metadata and project files.

Software & Tools

• Python: Cleans metadata, joins tables, and runs the correlation analysis.
• Jupyter Notebook: Keeps code, notes, and plots in one shareable file.
• Scanpy: Filters single-cell matrices, labels cell types, and makes quick plots.
• NetworkX: Builds metabolite-gene graphs and measures module structure.
• Seaborn: Makes heat maps and comparison plots for disease and control groups.

Experiment Steps

1. Define one disease comparison, one tissue source, and one cell type before you touch the data.
2. Choose a single public dataset set or a small matched set so your samples line up cleanly.
3. Decide how you will normalize each data type before you calculate any correlations.
4. Build a baseline network first, then define your new statistic for module strength, stability, or separation.
5. Set controls that test whether your links survive shuffling, resampling, and batch correction.
6. Plan how you will compare early Alzheimer's modules across donors, cell types, or brain regions.

Common Pitfalls

• Mixing datasets from different brain regions or assay platforms, which can turn batch effects into fake co-regulation.
• Treating a strong metabolite-gene correlation as proof of causation.
• Keeping too many low-quality single-cell features, which fills the network with dropout noise.
• Pairing transcriptomics and metabolomics tables after the labels are mismatched, which scrambles your modules.
• Building one giant network without splitting by cell type, which can hide changes that only show up in specific cells.

What Makes This Competitive

A class-level version stops at one network and a pretty plot. A stronger project tests whether your statistic still works after resampling, batch correction, and donor splits. You can also compare cell types or brain regions instead of averaging everything together. That gives you a real method question, not just a disease summary.

Project Variations

• Compare neurons, microglia, and astrocytes to see whether each cell type shows a different metabolite-gene module pattern.
• Swap Pearson correlation for partial correlation or mutual information, then compare which edges survive disease filtering.
• Test whether early Alzheimer's modules differ across brain regions, donor ages, or APOE risk groups in public datasets.

Learn More

• Human Cell Atlas Data Portal: Search public single-cell datasets and metadata for healthy and disease studies on the Human Cell Atlas website.
• Broad Institute Single Cell Portal: Browse downloadable single-cell studies, matrices, and annotations in a free public portal.
• NCBI GEO: Find transcriptomics and multi-omics studies by searching GEO for Alzheimer's and single-cell terms.
• PubMed: Search review articles on single-cell transcriptomics, metabolomics, and Alzheimer's network analysis.
• MIT OpenCourseWare: Find free lectures on statistics, probability, and data analysis when you need a refresher.