Machine Learning-Guided Mapping Sleep-Promoting Volatiles in Aromatic Plants

Project Description

This repository provides a machine-learning pipeline for identifying sleep-promoting volatile organic compounds (VOCs) from aromatic plants, including pretrained base models and a stacking predictor for quick inference.

Dependency

The code has been tested in the following environment:

How to Use

Installation

Option A: From PyPI (simplest)

pip install sleep-model

Option B: Conda environment (recommended for RDKit/DeepChem)

conda env create -f environment.yaml -n sleep_model
conda activate sleep_model
# install this project from source (editable or regular)
python -m pip install -e .
# python -m pip install .

Option C: uv (fast installer; lockfile included)

python -m pip install uv
py -m venv .venv
.\.venv\Scripts\Activate.ps1  # Windows PowerShell
uv sync

File Structure

├── data/                   # Input data files
├── data_analysis/          # Data processing and analysis
├── models/                 # Pretrained base model files for Stacking model training
│   ├── RF/
│   │   ├── rf_MACCSkeys_random_0.ipynb
│   │   ├── rf_RDkit_random_0.ipynb
│   ├── SVM/
│   │   ├── svm_MACCSkeys_random_3.ipynb
│   ├── XGB/
│   │   ├── xgb_MACCSkeys_random_0.ipynb
│   │── stacking_predict.ipynb
├── predict_smiles.py 
└── README.md

These four models (rf_MACCSkeys, rf_RDkit, svm_MACCSkeys, xgb_MACCSkeys) are the base models that we use to train the final stacking model.

Predicting

Command-line (console script)

After installation, a console command is available:

sleep-predict --smiles "CC(=O)OC1=CC=CC=C1C(=O)O"

As a Python module

python -m predict_smiles --smiles "CC(=O)OC1=CC=CC=C1C(=O)O"

Batch prediction from CSV

Predict for a CSV file containing a SMILES column (default column name: smiles):

python predict_smiles.py --csv example/input.csv --out example/preds.csv

Customize the SMILES column name and encoding when needed (e.g., column SMILES):

python predict_smiles.py --csv example/input.csv --out example/preds.csv --smiles-column SMILES --input-encoding utf-8

Notes

• Models and training data are loaded from the installed package resources (project models/ and data/GABAA.csv). Ensure they are present if running from source.
• If the console command is not found on Windows, re-activate your environment or run the module form.

Troubleshooting

• RDKit/DeepChem wheels can be environment-specific. If installation via pip fails, prefer the Conda-based installation.
• If you see a file-not-found error for models/ or data/GABAA.csv, run from the project root or install the project so resources are available in the environment.