ADME prediction with KNIME: In silico aqueous solubility consensus model based on supervised recursive random forest...

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• Accurate in-silico prediction of aqueous solubility is crucial for efficient drug discovery and development.
• Historically, poor model performance was attributed to data quality, but recent evidence suggests a need for improved methodologies.
• Existing solubility prediction models often lack the precision required for pharmaceutical applications.

Purpose of the Study:

• To develop novel machine learning approaches for enhanced aqueous solubility prediction.
• To create a robust and automated workflow for solubility prediction using curated data.
• To benchmark a new consensus model against existing literature methods.

Main Methods:

• Generation of a large, diverse aqueous solubility database from public sources.
• Development of two recursive machine learning approaches for data cleaning and variable selection.
• Construction of a consensus model integrating regression and classification algorithms.
• Implementation of the modeling protocol in KNIME for an automated prediction workflow.

Main Results:

• A comprehensive and curated database of aqueous solubility values was established.
• Novel recursive machine learning techniques were successfully applied for data preprocessing.
• The developed consensus model achieved results comparable to or better than existing literature models.
• An automated workflow for in-silico solubility prediction was successfully implemented.

Conclusions:

• The developed consensus model represents a significant advancement in in-silico aqueous solubility prediction.
• The study highlights the importance of advanced algorithms and data curation for accurate solubility modeling.
• The automated KNIME workflow facilitates efficient prediction for new chemical entities in drug discovery.