ScanFold2.0

ScanFold2.0 identifies RNA secondary structures with unusually stable folding by comparing native RNA sequences against randomized controls using machine learning-based z-score calculations.

Installation

Prerequisites

• Python 3.10 or higher
• Conda (recommended) or pip

Quick Setup with Conda

# Clone repository
git clone https://github.com/moss-lab/ScanFold2.0.git
cd ScanFold2.0

# Create environment
conda env create -f environment.yml
conda activate scanfold2

Alternative Setup with pip

macOS (Apple Silicon):

python3 -m venv scanfold_env
source scanfold_env/bin/activate
pip install -r requirements-mac.txt

Linux/Windows:

python3 -m venv scanfold_env
source scanfold_env/bin/activate  # Windows: scanfold_env\Scripts\activate
pip install -r requirements-linux.txt

Note: ViennaRNA must be installed separately via conda or system package manager.

Input Requirements

• Format: FASTA file containing RNA sequence(s)
• Sequence type: RNA sequences (U's) or DNA (T's will be converted to U's)
• Length: Sequences should be longer than window size (default 120 nt)
• Multiple sequences: Each sequence in the FASTA will be analyzed separately

Quick Start

# Basic usage
python ScanFold2.0.py input.fasta

# With custom parameters
python ScanFold2.0.py input.fasta -w 120 -s 1 -t 37 --shuffle di

Parameters

Parameter	Description	Default
-w	Window size (nt)	120
-s	Step size (nt)	1
-t	Temperature (°C)	37
--shuffle	Shuffling type: mono or di	mono
--folder	Output folder name	ScanFold_YYYYMMDD_HHMMSS

Shuffling Types

• mono: Mononucleotide shuffling - preserves single nucleotide composition
• di: Dinucleotide shuffling - preserves dinucleotide frequencies for better background modeling

Output Files

All results are saved to a timestamped folder (or custom named folder with --folder). Key outputs include:

• .tsv: Main results table with z-scores for each window
• .ct files: RNA structures in CT format (filtered at different z-score thresholds)
• .dbn files: Dot-bracket notation structures
• .wig files: Track files for genome browser visualization
• RESULTS_README.md: Detailed guide for viewing and interpreting results
• ScanFold_run.log: Complete analysis log

Examples

Basic RNA structure scanning

python ScanFold2.0.py my_rna.fasta

High-resolution scanning with dinucleotide shuffling

python ScanFold2.0.py my_rna.fasta -w 80 -s 1 --shuffle di --folder my_analysis

Temperature-specific folding

python ScanFold2.0.py viral_rna.fasta -t 30 -w 150

Interpreting Results

• Z-score < -1: Potentially stable structures
• Z-score < -2: Highly stable, likely functional structures
• Lower z-scores indicate more stable folding compared to random sequences

Visualizing in IGV

The WIG and BP files can be loaded into IGV (Integrative Genomics Viewer) for visualization:

1. Download IGV from https://igv.org if not already installed

2. Load your reference genome or sequence file

3. Load ScanFold tracks:

   • File → Load from File
   • Navigate to your ScanFold output folder
   • Select the WIG files:
     • .scan-zscores.wig - Z-score track
     • .scan-MFE.wig - Minimum free energy values
     • .scan-ED.wig - Ensemble diversity values
     • .Zavg.wig - Average Z-scores per base pair

4. Customize visualization:

   • Right-click tracks to adjust color schemes
   • Set z-score track to use blue-white-red gradient (blue for negative/stable)
   • Adjust track height for better visibility

Troubleshooting

Common Issues

TensorFlow/Metal warnings on Mac

• These are normal and can be ignored
• The tool will still run correctly

Memory issues with large sequences

• Try increasing step size (-s) to reduce number of windows
• Consider splitting very long sequences

ViennaRNA not found

• Ensure ViennaRNA is installed: conda install -c bioconda viennarna
• Check it's accessible: RNAfold -h

NumPy version conflicts

• Must use NumPy < 2.0 for TensorFlow compatibility
• Check with: pip show numpy

Performance

• Runtime: ~1-2 seconds per 1000 nt (depending on step size)
• Memory: Generally < 2GB for typical sequences
• GPU: Will use Apple Metal on Mac or CUDA on Linux if available

Citation

If you use ScanFold2.0, please cite:

Andrews RJ, Rouse WB, O'Leary CA, Booher NJ, Moss WN. (2023). ScanFold 2.0: a rapid approach for identifying potential structured RNA targets in genomes and transcriptomes. PeerJ 10:e14361. https://doi.org/10.7717/peerj.14361

Support

For issues or questions, please open an issue on GitHub