K-MELLODDY

Overview

This repository contains a Python-based preprocessing pipeline for K-MELLODDY standard data format, primarily designed for tasks in ADME/T prediction. The pipeline supports SMILES standardization, outlier detection, feature scaling, and label creation, making it suitable for classification and regression tasks. This repository will be updated when the K-MELLODDY standard data format is changed.

Features

• SMILES Standardization:
  • Removes salts, isotopes, and stereochemistry (optional).
  • Standardizes tautomeric forms and calculates molecular scaffolds.
• Label Processing:
  • Handles binary, categorical, and continuous labels.
  • Converts continuous labels into classification labels using thresholds.
  • Scales experimental values using StandardScaler.
• Outlier Detection:
  • Supports multiple methods for outlier detection, including IQR, Local Outlier Factor (LOF), One-Class SVM, and Gaussian Mixture Models (GMM).
• Task Validation:
  • Ensures the dataset aligns with the selected task (classification or regression).
• Customizable Parameters:
  • Options to retain stereochemistry, remove salts, detect outliers, and handle duplicates.

Installation

1. Clone the repository:

   git clone https://github.com/HITS-AI/k-melloddy.git
   cd k-melloddy

2. Install the required dependencies: (will be added soon)

   pip install -r requirements.txt

Usage

Example Code

from preprocessor import Preprocessor

# Initialize the Preprocessor
preprocessor = Preprocessor(
    input_path='data/chemical_data.csv',
    task='classification',
    task_name='solubility',
    smiles_column='SMILES_Structure_Parent',
    activity_column='Measurement_Value',
    remove_salt=True,
    keep_stereo=False,
    keep_duplicates=False,
    detect_outliers=True,
    threshold=50
)

# Run Preprocessing
processed_data = preprocessor.preprocess()

# Save the processed data
processed_data.to_csv('data/processed_chemical_data.csv', index=False)

Input File Format

The input file should be a CSV file with at least two columns:

• SMILES Column: Contains SMILES strings of compounds (default: SMILES_Structure_Parent).
• Activity Column: Contains numeric or categorical activity values (default: Measurement_Value).

Parameters

Parameter	Type	Description
input_path	str	Path to the input CSV file.
task	str	Task type: classification or regression.
task_name	str	Name of the task (e.g., solubility, cyp1a2 inhibition).
smiles_column	str	Name of the column containing SMILES strings.
activity_column	str	Name of the column containing activity values.
remove_salt	bool	Whether to remove salts from SMILES strings.
keep_stereo	bool	Whether to retain stereochemistry in SMILES strings.
keep_duplicates	bool	Whether to keep duplicate entries.
detect_outliers	bool	Whether to detect and remove outliers.
threshold	float	Threshold for converting continuous labels into classification labels (required for classification tasks).

Methods

Key Methods

• preprocess(): Runs the complete preprocessing pipeline.
• preprocess_compound(smiles): Processes a single SMILES string.
• detect_outliers_statistical(): Identifies outliers using the Interquartile Range (IQR) method.
• detect_outliers_density_based(): Identifies outliers using Local Outlier Factor (LOF).
• detect_outliers_classification_based(): Identifies outliers using One-Class SVM.
• scale_experiment_values(labels): Scales numeric activity values.

Dependencies

• pandas
• numpy
• rdkit
• scikit-learn
• scipy

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgments

• RDKit: For chemical informatics and machine learning tools.
• scikit-learn: For machine learning algorithms and preprocessing utilities.

Contact

For questions or issues, please open an issue on the repository or contact the maintainer at [email protected].