Flame: an open source framework for model development, hosting, and usage in production environments.

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• Predictive modeling is crucial for assessing drug candidate safety.
• Existing tools may lack flexibility or integration for production environments.
• The pharmaceutical industry requires robust, customizable modeling solutions.

Purpose of the Study:

• To introduce Flame, an open-source software for building and deploying predictive models.
• To provide a flexible and user-friendly platform for cheminformatics and machine learning workflows.
• To support collaborative model development and ensure model reproducibility.

Main Methods:

• Flame offers a web-based graphical interface for model building and customization.
• Supports structural normalization, molecular descriptor calculation, and machine learning model generation.
• Includes predefined and customizable workflows using state-of-the-art algorithms.
• Features a model repository with versioning and unique model IDs for documentation.

Main Results:

• Flame enables the creation of predictive models from biologically annotated chemical structures.
• The software allows customization of normalization, descriptors, and machine learning algorithms.
• Models can be exported for collaborative development and are stored with detailed documentation.
• Developed over three years with pharmaceutical industry collaboration.

Conclusions:

• Flame provides a comprehensive solution for building and deploying predictive models in research and industry.
• Its open-source nature and customizable workflows enhance collaborative drug discovery and safety assessment.
• The software addresses the need for robust, documented, and version-controlled predictive models.