New AI-Powered Tools
MolGenAI – Generative Chemistry:
- AI-driven molecule generation using REINVENT4-based technology (trained on ChEMBL and other datasets).
- Users can focus the generation on specific chemical space via transfer learning — refining models using example compounds.
- Integrates with the Radial Plot scoring to filter generated molecules based on target properties.
Flare AI Assistant:
- Chat model: answers questions about features, workflows, and underlying scientific methods.
- Code Assistant (Copilot): helps write and customize Python scripts using natural language, runnable directly in Flare.
Enhanced Modelling Capabilities
Protein-Protein Docking with Constraints:
- Users can now impose spatial constraints to bias docking — useful for modeling specific orientations (e.g., ternary complexes).
Detached Calculations:
- Long-running compute jobs (via Cresset Engine Broker v3.5) can run independently of the Flare GUI.
- You can close the interface while jobs continue on an HPC cluster, improving flexibility and resource management.
New Scoring & Analysis Features
Prevalence Score:
- Estimates how common molecular fragments are in curated databases (ChEMBL, eMolecules) as a proxy for synthetic accessibility.
- Fragments are scored from 0 (rare/complex) to 1 (common/simple), and the lowest score in a molecule defines its overall prevalence.
- Color-by-score visualization highlights problematic regions.
Performance & Data Handling
- Big improvements under the hood for handling very large projects (tens of thousands of ligands, hundreds of proteins).
- More efficient memory use and responsiveness.
Nested Ligand Roles:
- Supports parent/child relationships among molecular variants (e.g., tautomers, protonation states) in the ligand table for better organization.
FEP & Dynamics Enhancements
Flare FEP:
- Faster graph generation and improved handling of peptides.
- Support for machine-learned OpenFF partial charges (via AshGC) to speed parameterization for large molecules.
- New FEP Overview tab consolidates predicted vs experimental ΔG and structure data.
- Absolute FEP restraints and plot adjustments now better tuned and presented.
Dynamics:
- AMBER force field includes improved support (e.g., ZAFF for zinc).
- Trajectories can be exported in widely compatible formats (XTC).
- Better controls for RMSD plots and PCA on trajectory subsets.