RANQSAR: a standalone open-source application for reproducible machine learning-based QSAR analysis

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Quantitative structure-activity relationship (QSAR) modeling is vital in drug discovery.
• Developing reproducible QSAR models often requires complex workflows and multiple software tools.
• Standardized, user-friendly QSAR development environments are needed.

Purpose of the Study:

• Introduce RANQSAR, an open-source desktop application for reproducible machine learning-based regression QSAR modeling.
• Provide a unified graphical interface for the entire QSAR workflow, from data preparation to prediction.
• Facilitate standardized QSAR practices and comparative benchmarking.

Main Methods:

• RANQSAR integrates descriptor calculation (Morgan fingerprints, RDKit physicochemical descriptors), dataset splitting, model construction, and validation.
• Includes cross-validation, external validation, y-randomization, and applicability domain analysis for robust model assessment.
• Offers deterministic execution through seeding for reproducible results.

Main Results:

• Demonstrated the development of a reproducible regression QSAR model using acetylcholinesterase inhibitors as a case study.
• Showcased RANQSAR's capability for comparative benchmarking of different molecular representations and algorithms.
• Validated the platform's effectiveness in ensuring proper QSAR practices.

Conclusions:

• RANQSAR provides a practical, user-friendly, and reproducible environment for QSAR model development.
• It is a valuable tool for academics, students, and experts in computational drug discovery.
• The open-source nature and comprehensive documentation facilitate its adoption and use.