Orphan GPCR Pocket Clustering

ISEF Category: Biochemistry

Subcategory: Structural Biochemistry  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

Some human receptors still do not have a known natural partner. That gives you a real research puzzle, not a classroom exercise. You can use predicted protein shapes to sort orphan GPCRs by pocket similarity, then compare those groups with human metabolites. The result is a testable map of which molecules might fit where.

What Is It?

Think of a GPCR as a lock in the cell membrane. An orphan GPCR is a lock with no confirmed key. AlphaFold gives you a predicted 3D shape for the receptor, and pocket clustering groups receptors by the shape and chemistry of the cavity where a ligand might bind.

That matters because similar pockets often like similar ligands. HMDB, the Human Metabolome Database, lists many small molecules your body makes or uses. If an orphan receptor clusters near receptors with known ligands, you can make a stronger guess about which metabolites deserve a closer look.

Why This Is a Good Topic

This topic works well because you can test it with public data, clear measurements, and repeatable code. You are not guessing from vibes, you are comparing pocket features, cluster patterns, and metabolite matches. The project connects protein structure, cell signaling, and metabolite biology, so it has real scientific weight. You can learn how researchers turn raw structure data into ranked biological hypotheses.

Research Questions

• How does pocket similarity group orphan GPCRs relative to receptors with known endogenous ligands?
• What is the effect of using different pocket descriptors, such as volume, polarity, and residue composition, on the clustering result?
• Does filtering AlphaFold models by confidence score change which orphan receptors look most similar to known-ligand receptors?
• To what extent do HMDB metabolites match the top-ranked pocket clusters for each orphan receptor?
• Which orphan GPCRs stay in the same cluster after bootstrap resampling of the feature set?
• How does family-level grouping, such as class A versus class B GPCRs, change the ligand candidate list?

Basic Materials

• Computer with internet access and enough storage for structure files
• Python 3 installed locally or in Google Colab
• Jupyter Notebook
• Spreadsheet software
• AlphaFold Protein Structure Database access
• UniProt access
• HMDB access
• Free molecular viewer such as Mol* or UCSF ChimeraX

Advanced Materials

• Workstation with high-memory CPU or GPU access
• Unix shell environment
• Python scientific stack
• Biopython or MDAnalysis
• Pocket detection tool such as fpocket
• Clustering and statistics tools in Python or R
• Molecular docking suite such as AutoDock Vina
• Access to a university compute cluster
• Curated metabolite library from HMDB

Software & Tools

• Python: Cleans protein features, runs clustering, and makes plots for your receptor groups.
• Jupyter Notebook: Keeps your code, notes, and figures together in one place.
• Mol*: Lets you inspect predicted receptor structures in the browser and compare pockets.
• UCSF ChimeraX: Helps you visualize residue contacts, pocket shape, and model confidence.
• RStudio Desktop: Runs clustering checks and creates publication-style figures.

Experiment Steps

1. Define the receptor set you will study, including orphan GPCRs and a matched set of receptors with known endogenous ligands.
2. Choose the pocket features you will measure, such as volume, polarity, residue mix, and confidence score.
3. Build a clean reference table from AlphaFold models and any known structures you can use as controls.
4. Cluster the receptors and check whether orphan GPCRs group near receptors with similar ligand chemistry.
5. Rank HMDB metabolites against the strongest pocket clusters and decide how you will score matches.
6. Plan a validation test with negative controls so you can see whether your ranking beats random chance.

Common Pitfalls

• Using AlphaFold models with very low confidence, which can make weak pockets look real.
• Mixing receptor families without matched controls, which can blur the cluster pattern.
• Relying on one pocket scoring tool, which makes the result depend on a single algorithm.
• Treating HMDB presence as proof of binding, which confuses a metabolite list with receptor specificity.
• Skipping feature scaling before clustering, which lets one measurement dominate the full analysis.

What Makes This Competitive

A stronger project goes beyond a simple cluster plot. You compare orphan receptors against known-ligand GPCRs, then test whether the same pairing still appears after confidence filtering, feature scaling, and bootstrap resampling. You also add negative controls so you can estimate how often the pattern appears by chance. That kind of analysis turns a database search into a serious prediction pipeline.

Project Variations

• Compare orphan GPCR pockets against only peptide receptors, then see whether peptide-like metabolites cluster closer than lipid-like ones.
• Swap pocket clustering for ligand fingerprint matching and test whether both methods rank the same orphan receptors near the top.
• Limit the study to class A GPCRs and ask whether family-specific clustering gives cleaner ligand predictions.

Learn More

• AlphaFold Protein Structure Database: Search predicted human protein structures by gene or UniProt ID.
• UniProt: Find curated receptor sequences, names, and functional notes for GPCRs.
• HMDB: Look up human metabolites, classes, and structural records in the Human Metabolome Database.
• PubMed: Search review articles on orphan GPCRs, ligand deorphanization, and pocket analysis.
• RCSB PDB: Compare your predicted receptors with known GPCR structures used as controls.
• NCBI Bookshelf: Read free background chapters on membrane proteins, receptors, and protein structure.