Flame

Flame constructs predictive models from biologically annotated chemical structures and manages their versioned storage to support drug candidate safety assessment.

Key Features:

• Structural normalization: Performs structural normalization of chemical structures and provides selectable normalization techniques.
• Molecular descriptor generation: Generates molecular descriptors from normalized chemical structures and supports a range of descriptors implemented natively.
• Machine learning model construction: Builds machine learning models from molecular descriptors using a variety of algorithms implemented within the framework.
• Predefined workflows: Supplies predefined computational workflows for descriptor calculation and model building that can be customized.
• Extensibility: Allows modification of source code to enable unlimited model customization.
• Persistent repository and versioning: Stores models in a persistent repository with versioning and unique model identifiers.
• Implementation: Implements computational components on a Python backend.

Scientific Applications:

• Drug candidate safety assessment (IMI eTRANSAFE): Used to develop predictive models for assessing drug candidate safety using shared legacy data within the IMI eTRANSAFE project.
• Pharmacological research and development: Supports building and validating predictive models from biologically annotated chemical structures for pharmacological R&D.

Methodology:

Input are biologically annotated chemical structures; computational steps include structural normalization (selectable techniques), molecular descriptor generation, and machine learning model construction using native algorithms, orchestrated via predefined customizable workflows, with models stored in a versioned repository; components are implemented in a Python backend.