guliya_project

Guliya Glacier Virome — Dark Proteome & Novel Fold Discovery

🧊 Guliya Glacier Virome — Dark Proteome & Novel Fold Discovery

Overview

This repository contains a fully computational pipeline to:

1. Call genes from Guliya glacier vOTUs (viral Operational Taxonomic Units)
2. Annotate proteins using structure-informed phage annotation (Phold)
3. Extract the dark fraction — proteins with no known function
4. Cluster dark proteins to remove redundancy
5. Predict 3D structures using ESMFold
6. Screen for novel folds using Foldseek against PDB, AFDB50, and SCOP
7. Validate structural novelty using DALI and ECOD

Dataset: 1,951 Guliya glacier vOTUs assembled from ancient ice core metagenomes
Key result: 55.1% of all predicted proteins (17,169 / 31,135) are functionally dark

---

Pipeline Summary

Input: guliya_vOTUs.fasta (1,951 contigs) ↓ [Step 3] prodigal-gv gene calling → 31,135 proteins ↓ [Step 4] Phold structure-informed annotation ↓ [Step 5] Extract dark fraction → 17,169 dark proteins (55.1%) ↓ [Step 6] MMseqs2 clustering (90% identity) → 15,524 representative sequences ↓ [Step 7] Length filter (<= 1000 aa) → 15,437 ESMFold-ready proteins ↓ [Step 8] ESMFold structure prediction → 15,437 PDB files ↓ [Step 9] pLDDT quality filter (>= 60) ↓ [Step 10] Foldseek novel fold screening → Novel fold candidates ↓ [Step 11] DALI / ECOD structural validation ↓ [Step 12] DeepFRI functional inference

text

---

Repository Structure

Guliya_project/ ├── README.md ├── LICENSE ├── CHANGELOG.md ├── .gitignore ├── envs/ │ ├── guliya.yml │ ├── pholdENV.yml │ └── defensefinder.yml ├── scripts/ │ ├── install_tools.sh │ ├── step2_check_input.sh │ ├── step3_prodigal.sh │ ├── step4_phold.sh │ ├── step5_extract_dark.py │ ├── step6_cluster.sh │ ├── step7_length_filter.py │ ├── step8_run_esmfold.py │ ├── step9_plddt_filter.py │ └── step10_foldseek.sh ├── input/ # Place guliya_vOTUs.fasta here (not tracked by git) ├── gene_calls/ # prodigal-gv output ├── annotation/ # Phold output ├── structures/ # ESMFold predicted PDBs ├── novel_folds/ # Foldseek hits TM-score < 0.5 └── results/ # Final tables and summaries

text

---

Quick Start

1. Clone and set up environments

git clone https://github.com/<your-username>/Guliya_project.git
cd Guliya_project
bash scripts/install_tools.sh

2. Place your input file

cp /path/to/guliya_vOTUs.fasta input/

3. Run the pipeline step by step

conda activate guliya

bash scripts/step2_check_input.sh
bash scripts/step3_prodigal.sh
bash scripts/step4_phold.sh
python scripts/step5_extract_dark.py
bash scripts/step6_cluster.sh
python scripts/step7_length_filter.py
python scripts/step8_run_esmfold.py   # Run on GPU cluster
python scripts/step9_plddt_filter.py
bash scripts/step10_foldseek.sh

---

Key Results

Metric	Value
Input vOTUs	1,951
Total proteins predicted	31,135
Annotated by Phold	13,966 (44.9%)
Dark fraction	17,169 (55.1%)
After 90% identity clustering	15,524
ESMFold-ready (≤ 1000 aa)	15,437
Skipped (> 1000 aa)	87

---

Tool Requirements

Tool	Purpose
prodigal-gv	Phage-aware gene calling
Phold	Structure-informed annotation
MMseqs2 ≥ 14	Protein clustering
ESMFold (esm ≥ 2.0)	Structure prediction
Foldseek	Structural similarity search
Biopython ≥ 1.81	Sequence I/O
pandas ≥ 2.0	Data manipulation

---

Hardware Notes

• Gene calling & clustering: Runs on any machine (Mac, Linux)
• Phold: GPU-accelerated via Apple MPS on M-series Macs or CUDA on Linux
• ESMFold: Requires NVIDIA CUDA GPU (≥ 10GB VRAM) or HPC cluster with LocalColabFold

---

Citation

If you use this pipeline, please cite:

• Phold: Larralde et al., 2024
• ESMFold: Lin et al., 2023, Science
• Foldseek: van Kempen et al., 2024, Nature Biotechnology
• MMseqs2: Steinegger & Söding, 2017, Nature Biotechnology
• prodigal-gv: Camargo et al., 2023

---

License

MIT — see LICENSE for details.