Automated Workflows for Data Curation and Machine Learning to Develop Quantitative Structure-Activity Relationships.

Area of Science:

• Computational Chemistry
• Chemoinformatics
• Machine Learning

Background:

• Advancements in machine learning and large chemical datasets drive interest in computational chemistry tools.
• Quantitative Structure-Activity Relationship (QSAR) modeling is crucial for drug discovery and chemical research.

Purpose of the Study:

• To present a standard procedure for developing QSAR models.
• To implement two accessible and freely available workflows for QSAR model development and application.

Main Methods:

• Workflow 1: Data retrieval (SMILES) from the web, data validation, and curation for cheminformatics.
• Workflow 2: Implementation of six machine learning algorithms for classification QSAR model development.
• Inclusion of chemical descriptor calculation, hyperparameter tuning, and methods for handling data imbalance.

Main Results:

• Developed two integrated workflows in KNIME for QSAR model development.
• Enabled users to retrieve, curate, and analyze chemical data efficiently.
• Facilitated the creation of predictive QSAR models using various machine learning techniques.

Conclusions:

• The developed KNIME workflows provide a valuable tool for computational scientists.
• These workflows offer an intuitive introduction to QSAR modeling for researchers.
• The tools support the prediction of activities for external chemical compounds.