Prion Prediction Pipeline

A comprehensive R-based pipeline for identifying prion-like domains in fungal proteomes. This pipeline analyzes UniProt fungi databases to predict proteins with prion-forming propensity using multiple complementary algorithms.

Overview

The pipeline processes UniProt SwissProt/TrEMBL fungi databases through a series of analysis steps:

1. Data Pre-Analysis - Converts UniProt .dat files to tabular format, extracts taxonomy, and maps ecological guilds via FunGuild
2. PrionScan - Predicts Q/N-rich prion domains using probabilistic scoring (Espinosa Angarica algorithm)
3. PAPA - Prion Aggregation Prediction Algorithm with FoldIndex filtering (Toombs et al.)
4. PLAAC - Prion-Like Amino Acid Composition analysis using log-likelihood scoring (Lancaster et al.)

Requirements

R Packages

install.packages(c("tidyverse", "FUNGuildR", "patchwork", "optparse", "parallel", "readr", "stringr", "ggVennDiagram"))

Java (for PLAAC)

# Option 1: System Java (JRE 8+)
java -version  # Verify installation

# Option 2: Conda environment
conda create -n plaac-java openjdk=11 -c conda-forge
# The script can auto-setup this with: Rscript R/plaac.R --setup-java

Git is also required for initial PLAAC repository clone.

Optional (for diagram generation)

# For PNG export via mermaid.ink API
install.packages("base64enc")

# OR local CLI (no internet required)
npm install -g @mermaid-js/mermaid-cli

Directory Structure

prion-prediction-pipeline/
├── R/                          # Analysis scripts
│   ├── pipeline.R              # Orchestrator (runs all steps)
│   ├── data_preanalysis.R      # Step 1: Data conversion & FunGuild
│   ├── PrionScan.R             # Step 2: Prion domain prediction
│   ├── PAPA.R                  # Step 2: Aggregation propensity
│   ├── plaac.R                 # Step 2: PLAAC prion-like composition
│   └── venn_predictions.R      # Compare all three methods
├── data/
│   ├── raw/                    # Input: UniProt .dat files
│   ├── cache/                  # Intermediate: Tabular cache, FunGuild DB
│   └── processed/              # Output: Predictions, taxonomy, guild mapping
├── docs/                       # Documentation & changelogs
│   ├── workflow.md             # Full workflow description
│   └── changelog_*.md          # Version history per script
└── reports/                    # Pipeline run logs

Quick Start

1. Download Input Data

Place UniProt fungi proteome files in data/raw/:

# Swiss-Prot (~50MB, curated)
wget -P data/raw/ https://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/taxonomic_divisions/uniprot_sprot_fungi.dat.gz

# TrEMBL (~73GB, comprehensive)
wget -P data/raw/ https://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/taxonomic_divisions/uniprot_trembl_fungi.dat.gz

2. Run Full Pipeline

# Process Swiss-Prot database (recommended for testing)
Rscript R/pipeline.R --db sprot

# Process TrEMBL database (large, requires significant time/memory)
Rscript R/pipeline.R --db trembl

3. Run Individual Steps

# Step 1: Data pre-analysis only
Rscript R/data_preanalysis.R --db sprot

# Step 2a: PrionScan only (requires Step 1 cache)
Rscript R/PrionScan.R --db sprot

# Step 2b: PAPA only (requires Step 1 cache)
Rscript R/PAPA.R --db sprot

# Step 2c: PLAAC only (requires Step 1 FASTA output)
Rscript R/plaac.R -i data/processed/sprot_fungi_sequences_*.fasta --db sprot

# Compare predictions (auto-detects all available methods)
Rscript R/venn_predictions.R --db sprot

Pipeline Orchestrator

The pipeline.R script coordinates all analysis steps:

# Show available steps and status
Rscript R/pipeline.R --list --db sprot

# Dry-run (validate inputs without executing)
Rscript R/pipeline.R --check --db sprot

# Run specific step only
Rscript R/pipeline.R --db sprot --step PrionScan

# Export workflow diagram
Rscript R/pipeline.R --diagram

Output Files

From data_preanalysis.R

File	Description
*_fungi_tabular_*.tsv	Cached protein data (ID, Organism, Taxonomy, TaxID, Sequence)
*_fungi_taxonomy_*.tsv	Protein taxonomy mapping
*_fungi_sequences_*.fasta	FASTA sequences (header: >ID|TaxID|Organism)
*_genus_guild_mapping_*.tsv	FunGuild ecological guild annotations

From PrionScan.R

File	Description
*_prion_predictions_*.tsv	Prion domain predictions (Score >= 50)

Columns: Protein_ID, Organism, Window_Position, Score, Prion_Domain

From PAPA.R

File	Description
*_papa_predictions_*.tsv	PAPA scores for all proteins

Columns: Protein_ID, Organism, PAPA_Score, PAPA_Position, Above_Threshold

From plaac.R

File	Description
*_plaac_full_*.tsv	Full 37-column PLAAC output (all scores and metrics)
*_plaac_predictions_*.tsv	Standardized predictions for pipeline integration

Full output includes: COREscore, LLR, PAPA, FoldIndex disorder, MW enrichment, sequence positions.

Standardized columns: Protein_ID, Organism, Prion_Score, LLR_Score, PAPA_Score, Has_Prion_Domain, Domain_Start, Domain_End, Domain_Length, Domain_Sequence, Protein_Length

From venn_predictions.R

File	Description
*_venn_predictions_*.png	Venn diagram comparing all available prediction methods (2-way or 3-way)

Note: Taxonomy and NCBI_TaxID are stored in the tabular cache. Join by Protein_ID when needed.

Workflow Diagram

flowchart TD
    DAT[("UniProt .dat")] --> S1[data_preanalysis.R]
    S1 --> TSV[("Tabular Cache")]
    S1 --> TAX[("Taxonomy TSV")]
    S1 --> FASTA[("FASTA Sequences")]
    S1 --> GUILD[("Guild Mapping")]
    TSV --> S2[PrionScan.R]
    TSV --> S3[PAPA.R]
    FASTA --> S4[plaac.R]
    S2 --> PRION[("PrionScan Predictions")]
    S3 --> PAPA_OUT[("PAPA Predictions")]
    S4 --> PLAAC_OUT[("PLAAC Predictions")]
    PRION --> VENN[venn_predictions.R]
    PAPA_OUT --> VENN
    PLAAC_OUT --> VENN
    VENN --> VENN_OUT[("Venn Diagram")]

Loading

Algorithm Details

PrionScan

• Method: Sliding window (60 aa) with amino acid propensity scoring
• Threshold: Score >= 50 for positive prediction
• Reference: Espinosa Angarica et al. (2013) BMC Genomics; Alberti et al. (2009) Cell

PAPA

• Method: Prion propensity with FoldIndex disorder filtering
• Window: 41 aa (configurable)
• Threshold: Score >= 0.05 for potential prion-forming propensity
• Reference: Toombs et al. (2010) Mol Cell Biol 30(1):319-332

PLAAC

• Method: Log-likelihood scoring for Q/N-rich sequences vs background proteome
• Core length: 60 aa minimum (configurable)
• Threshold: COREscore >= 20 for high-confidence hits
• Reference: Lancaster et al. (2014) BMC Bioinformatics; https://github.com/whitehead/plaac

Performance Notes

• Tabular caching: First run converts .dat to TSV (~4-5x faster). Subsequent runs use cache (~15-25x faster)
• Parallel processing: PrionScan and PAPA use all available cores by default (--cores N to customize)
• Memory: TrEMBL processing requires ~16GB+ RAM

Documentation

See docs/ for detailed documentation:

• workflow.md - Complete analysis workflow and tool descriptions
• changelog_*.md - Version history for each script

License

GNU GPL v3+

Citations

If you use this pipeline, please cite:

• PrionScan algorithm: Espinosa Angarica V, et al. (2013) "PrionScan: an online database of predicted prion domains in complete proteomes." BMC Genomics
• PAPA algorithm: Toombs JA, et al. (2010) "Compositional determinants of prion formation in yeast." Mol Cell Biol 30(1):319-332
• PLAAC algorithm: Lancaster AK, et al. (2014) "PLAAC: a web and command-line application to identify proteins with prion-like amino acid composition." Bioinformatics
• FunGuild: Nguyen NH, et al. (2016) "FUNGuild: An open annotation tool for parsing fungal community datasets by ecological guild." Fungal Ecology