NR-ToxPred

A desktop application and command-line tool for predicting the binding of chemical compounds to nine nuclear receptors (NRs) using pre-trained machine learning models.

---

Quick Start (No Python experience needed)

Step 1 — Download this repository Click the green Code button above → Download ZIP → Extract the ZIP folder anywhere on your computer

Step 2 — Install (one-time, ~10–20 minutes)

• Windows: Double-click install.bat
• Mac / Linux: Open a terminal in the folder and run bash install.sh

The installer automatically downloads Python (Miniconda) and all required packages. No prior setup needed.

Step 3 — Run the app

• Windows: Double-click run.bat
• Mac / Linux: Run bash run.sh

On first launch, click Download SVM only when prompted to fetch the prediction models (~250 MB).

Steps 1–2 are one-time only. After that, just use Step 3 every time.

---

Overview

NR-ToxPred predicts whether a compound is Active or Inactive against the following nuclear receptors:

Receptor	Full Name
RXR	Retinoid X Receptor
PR	Progesterone Receptor
GR	Glucocorticoid Receptor
AR	Androgen Receptor
ERA	Estrogen Receptor Alpha
ERB	Estrogen Receptor Beta
FXR	Farnesoid X Receptor
PPARD	Peroxisome Proliferator-Activated Receptor Delta
PPARG	Peroxisome Proliferator-Activated Receptor Gamma

Each prediction includes an Applicability Domain (AD) assessment — Reliable or Unreliable — based on Tanimoto fingerprint similarity to the training set.

---

Features

• GUI and CLI — interactive desktop app or scriptable command-line interface
• Single compound prediction — enter a SMILES string and get instant results
• Batch prediction — upload a CSV/Excel file with a column of SMILES strings
• Two fingerprint types — Morgan (ECFP6, 1024 bits) and MACCS Keys (167 bits)
• Two algorithms — SVM (fast, ~250 MB) and SuperLearner (ensemble, ~12 GB)
• Applicability Domain — Tanimoto-based AD with adjustable similarity cutoff and neighbor count
• 2D structure viewer — renders the molecule structure in the single prediction tab
• Molecular descriptors — MW, LogP, HBD, HBA, TPSA, RotBonds displayed per compound
• Export results — save batch predictions to Excel or CSV
• Auto-download — fetches models from Hugging Face Hub on first run

---

Command-Line Interface

NR-ToxPred can be used without the GUI, which is useful for scripting and headless servers.

Single compound:

python pytox_gui.py --no-gui --smiles "CC(=O)Oc1ccccc1C(=O)O" --name Aspirin

Batch from CSV:

python pytox_gui.py --no-gui --csv compounds.csv --smiles-col SMILES --output results.xlsx

Key options:

Option	Default	Description
--smiles SMILES	—	SMILES string (single compound)
--csv FILE	—	CSV or Excel file (batch)
--smiles-col COL	SMILES	Column name containing SMILES
--name NAME	Compound	Label for the compound
--fp {morgan,maccs}	morgan	Fingerprint type
--algo {svm,superlearner}	svm	Prediction algorithm
--receptors R [R ...]	all nine	Subset of receptors to predict
--scutoff FLOAT	0.25	AD Tanimoto similarity cutoff
--nsimilar INT	1	AD minimum similar neighbours
--output FILE	stdout	Output file (.csv or .xlsx)

Run python pytox_gui.py --help for the full option list.

---

Requirements

System dependencies (install via conda)

conda install -c conda-forge rdkit

Python packages

pip install -r requirements.txt

requirements.txt includes: molvs, scikit-learn==0.23.2, mlens, pandas, numpy, scipy, openpyxl, Pillow, huggingface_hub

Note: scikit-learn is pinned to 0.23.2 and mlens is required for SuperLearner models. Both are installed automatically by requirements.txt.

---

Model Files

The pre-trained models are not included in this repository due to their size. On first launch the app offers to download them automatically from Hugging Face Hub.

Model set	Size	Recommended for
SVM only	~250 MB	Most users — fast and accurate
SVM + SuperLearner	~12 GB	Maximum accuracy; large download

Download location:

• Windows: %LOCALAPPDATA%\NRToxPred\ (never synced by OneDrive)
• Mac / Linux: same folder as pytox_gui.py

Manual placement

If you prefer to copy model files yourself, place MODELS/ and X_train/ next to pytox_gui.py:

NRToxPred-GUI/
├── MODELS/
│   ├── morgan/
│   │   ├── ARsvm_best.model
│   │   └── ... (one per receptor)
│   ├── MACCS/
│   │   └── ... (one per receptor)
│   └── ARclasses.npy
└── X_train/
    ├── AR.xlsx
    └── ... (one per receptor)

---

Installation & Running (Python users)

# 1. Clone the repository
git clone https://github.com/gokulalgates/NRToxPred-GUI.git
cd NRToxPred-GUI

# 2. Create and activate a conda environment
conda create -n nrtoxpred python=3.8
conda activate nrtoxpred

# 3. Install RDKit via conda
conda install -c conda-forge rdkit

# 4. Install remaining dependencies
pip install -r requirements.txt

# 5. Launch the application
python pytox_gui.py

The app will prompt you to download models on first run.

---

Applicability Domain

Each prediction is tagged as:

• Reliable — the compound is similar to at least N training set compounds at a Tanimoto similarity ≥ S
• Unreliable — the compound falls outside the training set chemical space; predictions should be interpreted with caution

The Scutoff (similarity threshold) and Nsimilar (minimum neighbour count) parameters can be adjusted in the AD Parameters panel (GUI) or via --scutoff / --nsimilar (CLI).

---

Citation

If you use NR-ToxPred in your research, please cite:

Predicting the binding of small molecules to nuclear receptors using machine learning. Brief Bioinform. 2022 May 13;23(3):bbac114. doi: 10.1093/bib/bbac114

---

License

MIT License. See LICENSE for details.