🧪 ARDA is currently in beta — features may change. Send feedback →

Sign InSign Up
RNA-FOLD

RNA-FOLD

This guide prsqjkhdqsjkhdqsjkovides detailed explanations of all command-line interface (CLI) options and usage patterns for the main entry points of the project.

PythonPyTorchPyMol

Command-Line Interface and Usage Guide

This guide provides detailed explanations of all command-line interface (CLI) options and usage patterns for the main entry points of the project: main.py, main_lbfgs.py, and manage_dataset.py.

Optimizing 3D RNA Structures: main.py

The main.py script is the primary entry point to launch high-performance optimizations using Adam. It supports both interactive terminal Prompts and batch options suitable for cluster jobs.

How to Run

python main.py launch [OPTIONS]

General Options

  • --molecule [Protein|RNA]: The type of molecule to optimize (default: RNA).
  • --method [1|2]: The optimization method:
    • 1: Bead-springs (Coarse-Grained)
    • 2: Full atoms (All atoms)
  • --input-type [1|2]: The format of the input molecule:
    • 1: Direct sequence string (e.g., ACGU)
    • 2: Path to a FASTA file
  • --input-val <TEXT>: The sequence string or FASTA file path depending on the --input-type (Required in batch mode).
  • --output <TEXT>: The path to the output folder or specific filename where the optimized PDB structure will be saved.
  • --cif: Flag to export the final structure in mmCIF format in addition to PDB.
  • --verbose: Flag to enable detailed console logging.
  • --confirm: Flag to display a summary of configuration parameters and ask for user confirmation before launching.
  • --batch: Flag to run in non-interactive batch mode (suitable for automation/clusters).

Metrics and Evaluation

  • --target-structure <TEXT>: Path to a target/reference PDB structure to calculate comparative metrics (such as RMSD/MAE).
  • --chain <TEXT>: Chain ID to extract from the reference PDB structure.
  • --save-metrics: Flag to save the final optimization metrics to a CSV file.
  • --output-metrics <TEXT>: The filename for output metrics CSV (default: metrics.csv).

Visualization

  • --visualise: Flag to generate a 3D folding animation video (requires ffmpeg). Saves intermediate PNG snapshots every 200 epochs and builds an MP4 at the end.
  • --vis-metrics: Flag to save intermediate folding metrics (folding_vis.csv) without generating structural snapshots and the final MP4 video.
  • --vis-dir <TEXT>: Specific directory where visualization snapshots and metrics should be saved.

Optimization Parameters (Adam Engine)

  • --patience-locale <INTEGER>: Number of iterations without improvement (within --min-delta) before terminating a local optimization phase (default: 100).
  • --patience-globale <INTEGER>: Number of global cycles without improvement before stopping the overall Basin Hopping loop (default: 5).
  • --min-delta <FLOAT>: Minimum change in energy to qualify as an improvement (default: 1e-4).
  • --taux-refroidissement <FLOAT>: Cooling rate / decay factor for the noise base at each Basin Hopping cycle (default: 0.85).
  • --bruit-min <FLOAT>: Minimum coordinate noise in Angstroms under which global cycles stop (default: 0.01).
  • --noise-coords <FLOAT>: Initial coordinate noise perturbation factor in Angstroms (default: 1.5).
  • --bead-atom <TEXT>: The specific bead atom type representing each residue in coarse-grained optimization (default: C4').
  • --score <TEXT>: The statistical potential or scoring function to drive the optimization:
    • For RNA: RASP, DFIRE, rsRNASP, cgRNASP, RMSD, MAE, or All.
    • For Protein: DFIRE_P, RMSD, MAE, or All.
  • --score-weight <FLOAT>: Stochastic energy score multiplier weight (default: 1.0).
  • --epsilon-wca <FLOAT>: Repulsive scaling parameter for Weeks-Chandler-Andersen steric potential (default: 1.0).
  • --relax: Flag to physically relax the final all-atom PDB using OpenMM and Amber14 forcefield.

Full-Atom Specific Options

  • --backbone-weight <INTEGER>: Harmonic spring weight constraint to keep virtual backbone O3’-P covalent bonds intact (default: 100).
  • --clash-weight <INTEGER>: Steric clash penalty weight based on Van der Waals radii interpenetration (default: 100).
  • --noise-angles <FLOAT>: Initial angular noise perturbation factor in radians (default: 0.5).

L-BFGS Optimization: main_lbfgs.py

The main_lbfgs.py script exposes the exact same optimization suite but leverages the L-BFGS optimizer instead of Adam.

How to Run

python main_lbfgs.py launch [OPTIONS]

Additional L-BFGS Options

In addition to all options present in main.py, the following parameters are available:

  • --lr <FLOAT>: The learning rate / initial step size for L-BFGS (default: 0.2).
  • --init-structure <TEXT>: The path to an initial PDB structure to start the L-BFGS optimization from, enabling refinement of existing models.

Dataset and Statistics Management: manage_dataset.py

The manage_dataset.py script is a utility toolkit to download structures from RCSB PDB, generate distance/angle statistics, clean structures, or automate workflows.

It is structured into distinct subcommands.

How to Run

python manage_dataset.py [COMMAND] [OPTIONS]

Available Commands

retrieve_data

Downloads a dataset matching specific quality and resolution criteria.

python manage_dataset.py retrieve_data [FOLDER_PATH] [OPTIONS]
  • FOLDER_PATH (Argument): Directory where structures will be saved.
  • -m, --molecule [RNA|Protein]: Molecule type to download (default: RNA).
  • -r, --resolution FLOAT: Maximum allowed experimental resolution in Angstroms (default: 1.1).
  • -e, --extension [mmCif|pdb]: Target file format to download (default: mmCif).
  • --pure/--no-pure: Download only pure structures (e.g., only RNA, excluding complexes) (default: True).

make_distri

Extracts and saves consecutive residue distances from a folder of PDB/mmCIF structures.

python manage_dataset.py make_distri [FOLDER_PATH] [OPTIONS]
  • FOLDER_PATH (Argument): Folder containing structures to analyze.
  • -m, --molecule [RNA|Protein]: Molecule type (default: RNA).
  • -a, --ref-atom TEXT: The atom type to measure distances between (default: C3').
  • -o, --output TEXT: Prefix path for output CSV files (default: distances).
  • --save-distances/--no-save-distances: Save all individual pairwise measurements to a CSV (default: True).
  • --save-summary/--no-save-summary: Save overall statistical indicators (mean, std, min, max) to results_summary.csv (default: True).

extract_sequences

Extracts nucleotide sequences and headers from PDB files in a directory.

python manage_dataset.py extract_sequences [FOLDER_PATH] [OPTIONS]
  • FOLDER_PATH (Argument): Directory containing PDB files.
  • -o, --output TEXT: Output CSV filename (default: sequences.csv).

clean_pdb

Cleans PDB structures by stripping away protein chains and ligand heteroatoms to isolate pure nucleic acids.

python manage_dataset.py clean_pdb [FOLDER_PATH] [OPTIONS]
  • FOLDER_PATH (Argument): Directory containing PDB files.
  • -o, --output TEXT: Folder to save cleaned structures. If not specified, modifies files in-place.

auto

An interactive wizard that automates both dataset retrieval and statistical calculations.

python manage_dataset.py auto [FOLDER_PATH] [OPTIONS]
  • FOLDER_PATH (Argument): Target directory for dataset and statistics.
  • Accepts all options from retrieve_data and make_distri commands.