Possibilities for transfer of relevant data without revealing structural information

Area of Science:

• Computational chemistry
• Toxicology
• Drug discovery

Background:

• Early predictive ADME/Tox model development faced challenges in proprietary data sharing.
• Industry scientists needed to evaluate models but were hesitant to share chemical structures.
• Model developers could run predictions but not disclose key descriptors.

Purpose of the Study:

• To describe a secure data exchange process for predictive ADME/Tox model development.
• To enable model evaluation while protecting proprietary chemical structures and descriptors.
• To explore the feasibility of predictions without explicit structural knowledge.

Main Methods:

• Developed a process for data exchange using public descriptors.
• Scientists calculated properties and sent results as property files, not structures.
• Model developers used extracted descriptors to run predictions and return results.

Main Results:

• Successfully facilitated data exchange between scientists and model developers.
• Demonstrated a method to perform ADME/Tox predictions using descriptors only.
• Addressed concerns about compromising proprietary structural information via descriptors.

Conclusions:

• The described method allows for secure evaluation of predictive ADME/Tox models.
• ADME/Tox predictions can be made using descriptors without direct access to chemical structures.
• This approach balances the need for model validation with the protection of intellectual property.