Using open source computational tools for predicting human metabolic stability and additional absorption,...

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Ligand-based computational models are crucial for drug discovery but often require proprietary software, hindering data sharing.
• Open source alternatives offer a potential solution for more accessible and shareable computational modeling.

Purpose of the Study:

• To evaluate the performance of open source molecular descriptors and modeling algorithms against commercial software.
• To assess the feasibility of using open source tools for predicting metabolic stability, passive permeability, and P-glycoprotein efflux.

Main Methods:

• Open source descriptors (Chemistry Development Kit - CDK) and algorithms were compared with commercial software (Molecular Operating Environment 2D - MOE2D).
• Models were built using C5.0 decision trees and Cubist algorithms on datasets ranging from 25,000 to 193,000 molecules.
• Performance was evaluated using statistical metrics including kappa (κ), sensitivity, specificity, and positive predicted value (PPV).

Main Results:

• Open source CDK descriptors combined with SMARTS keys achieved statistical performance comparable to commercial MOE2D descriptors for predicting metabolic stability.
• Continuous models using CDK or MOE2D descriptors yielded similar predictive accuracy.
• Comparable model testing statistics were observed for passive permeability and P-glycoprotein efflux predictions.

Conclusions:

• Open source cheminformatics tools provide predictive results on par with commercial software, offering substantial cost savings.
• These open source tools can serve as a valuable asset for organizations to share data precompetitively, reducing redundancy and accelerating drug discovery efforts.