🧬 GREB1L Protein Structure Analysis

Comparative structural analysis of the GREB1-like protein (Q9C091) in Homo sapiens using AlphaFold2, Phyre2, and ESMFold.

---

📌 Table of Contents

• Overview
• Scientific Context
• Methods & Workflow
• Results Summary
• RMSD Analysis
• Quality Control Findings
• AlphaFold2 Advanced Metrics
• Project Structure
• How to Reproduce
• Tools & Technologies
• Contributors
• Conclusion

---

🔬 Overview

This project focuses on the comparative structural analysis of the GREB1-like protein (Q9C091) in Homo sapiens using three state-of-the-art protein structure prediction methods:

Method	Approach
AlphaFold2	Deep learning-based full structure prediction
Phyre2	Homology modeling using known templates
ESMFold	Protein language model-based prediction

The goal is to evaluate, compare, and validate predicted 3D protein structures using multiple bioinformatics and structural biology tools.

---

🧠 Scientific Context

RNA-related proteins and transcription-associated complexes often require accurate structural prediction to understand their biological function.

The GREB1-like protein:

• Contains 1923 amino acids
• Has no complete experimental structure available in PDB
• Requires computational modeling approaches

👉 This project aims to compare different prediction models, evaluate their reliability, and analyze structural consistency.

---

⚙️ Methods & Workflow

Protein Sequence (GREB1L — Q9C091)
              │
              ▼
┌──────────────────────────┐
│  1. Sequence Analysis    │──→ BLASTp homolog identification
└────────────┬─────────────┘
             │
     ┌───────┼───────┐
     ▼       ▼       ▼
┌────────┐┌────────┐┌────────┐
│Alpha-  ││Phyre2  ││ESM-    │
│Fold2   ││        ││Fold    │
└───┬────┘└───┬────┘└───┬────┘
    │         │         │
    └─────────┼─────────┘
              ▼
┌──────────────────────────┐
│  3. Structural Comparison│──→ RMSD calculation (PyMOL)
└────────────┬─────────────┘
             │
             ▼
┌──────────────────────────┐
│  4. Quality Assessment   │──→ ProQ2, Ramachandran, SAVES
└──────────────────────────┘
             │
             ▼
      ✅ Validated Models

---

1. Structure Prediction Models

🔹 AlphaFold2

• Deep learning-based model
• Predicts full protein structure
• Provides confidence scores (pLDDT, PAE)

🔹 Phyre2

• Homology modeling approach
• Uses known protein templates
• Confidence based on template similarity

🔹 ESMFold

• Language model-based prediction
• Faster but may predict partial structures

---

2. Sequence Analysis

BLASTp was used to identify homologous sequences. High similarity was found with GREB1-like isoforms (≈100% identity), confirming the biological relevance of the sequence.

---

3. Structural Comparison

Structures were compared using RMSD (Root Mean Square Deviation) to measure similarity between predicted models. Partial alignment was used for fair comparison across methods with different coverage.

---

4. Quality Assessment

Multiple tools were used for validation:

Tool	Purpose
ProQ2	Predicts global and local model quality
Ramachandran Plot	Evaluates backbone conformation (α-helices, β-sheets, outliers)
SAVES	Detects steric clashes, B-factor anomalies, side-chain issues

---

📊 Results Summary

Model	Domains Predicted	Coverage	Strength
AlphaFold2	4 domains	Full	Best global accuracy
Phyre2	1 domain (~228 aa)	Partial	Template-based, moderate reliability
ESMFold	1 small domain (~50–60 aa)	Very low	Best local precision

Key Insights

• AlphaFold2 → best global model with high confidence regions (pLDDT 60–90)
• ESMFold → best local precision but limited global coverage
• Phyre2 → template-dependent, less reliable, several structural inconsistencies

---

📐 RMSD Analysis

Comparison	RMSD (Å)	Interpretation
Phyre2 vs AlphaFold2	≈ 4.16	Moderate deviation, less accurate structurally
ESMFold vs AlphaFold2	≈ 2.81	Closer locally, higher structural agreement

👉 Lower RMSD = higher structural similarity. ESMFold shows better local agreement with AlphaFold2 than Phyre2.

---

🔍 Quality Control Findings

Common issues identified across models:

• Steric clashes (Van der Waals)
• Abnormal backbone conformations
• Side-chain optimization issues
• B-factor inconsistencies

---

🔵 AlphaFold2 Advanced Metrics

pLDDT (per-residue confidence):

• High values → reliable regions
• Low values (<50) → uncertain regions

PAE (Predicted Alignment Error):

• Shows domain positioning reliability
• Confirms stable domain regions

---

📂 Project Structure

greb1l-structure-analysis/
│
├── report/                 # PDF report and figures
│   └── final_report.pdf
│
├── figures/                # Structural visualizations
│   ├── alphafold2/         # AlphaFold2 PyMOL renders
│   ├── phyre2/             # Phyre2 model visualizations
│   ├── esmfold/            # ESMFold predictions
│   ├── ramachandran/       # Ramachandran plots
│   └── rmsd_comparison/    # Structural overlay figures
│
└── README.md               # Project documentation

---

🚀 How to Reproduce

1. Retrieve protein sequence:

• Download GREB1L (Q9C091) from UniProt

2. Run prediction tools:

• AlphaFold2
• Phyre2
• ESMFold

3. Perform analysis:

• BLAST analysis for homolog identification
• Structural alignment using PyMOL

4. Evaluate:

• RMSD calculation between models
• Ramachandran plot analysis
• ProQ2 quality scores

---

🧪 Tools & Technologies

Category	Tools
Structure Prediction	AlphaFold2, Phyre2, ESMFold
Sequence Analysis	BLAST+
Visualization	PyMOL
Quality Assessment	ProQ2, SAVES
Data Analysis	Python

---

👥 Contributors

• Fatine Hichami
• Tugce Koytaviloglu

---

🎯 Key Learnings

• Structural prediction models vary significantly in coverage and accuracy
• Combining multiple methods improves reliability of results
• Local vs global accuracy must be distinguished when evaluating models
• Quality control is essential in computational structural biology

---

📌 Conclusion

This project demonstrates that no single model is sufficient for protein structure prediction.

A combined approach using:

• AlphaFold2 for global structure
• ESMFold for local precision

provides a more robust understanding of protein conformation.