We are excited to announce the latest release of Flare V11, which brings a range of new features and significant enhancements designed to make computational drug discovery more efficient, accurate, and user-friendly. From new generative chemistry tools, improved Free Energy Perturbation (FEP) workflows and enhanced Protein-Protein Docking capabilities to smarter ligand management and more flexible calculation handling, this release addresses key challenges faced by researchers working with complex molecular systems.
Generate new molecules with MolGenAI
MolGenAI is the first exciting addition in this release: a powerful tool for generating new molecular structures using advanced AI techniques. Built on REINVENT4 technology1 and using priors trained in high-quality datasets such as ChEMBL, MolGenAI allows researchers to generate new molecules that explore entirely new areas of chemical space.
MolGenAI also offers transfer learning, which enables existing priors to be refocused on a narrower, user-defined region of chemical space using a small number of example compounds. This means you can tailor molecular generation to the novel compounds most relevant to your project. New, re-trained priors can be saved, reused and easily shared across the team.
To further streamline the selection of the best molecules, MolGenAI integrates seamlessly with Flare’s Radial Plot score, enabling filtering of the generated molecules based on how well they fit the desired physico-chemical profile. By adjusting the overall or individual component score thresholds, users can control the filtering, obtaining molecules that respect specified criteria.
In the example illustrated in the figure below we show how to generate new analogues of a known ligand. In this example, we start with the co-crystallized ligand from protein PDB: 1OIT.
- Initially, we assemble a set of structurally related compounds. Here we used R-group replacement from Spark™ in Flare, ensuring the key scaffold is preserved.
- Using the generated Spark molecules as a Training Set, we re-train the prior model so that it becomes biased toward the desired chemical space.
- Filtering is applied directly in the MolGenAI workflow: ranges/functions and weights for key properties (e.g., MW, SlogP, and Flexibility) can be set using the Radial Plot Properties widget. By setting a minimum component score of 0.7 for all three criteria, only molecules that closely match the design profile are retained.
- MolGenAI generates new molecules that are structurally aligned with our target series, match the desired property profile, while still offering diversity.
Work smarter and faster with the Flare AI Assistant
Another new feature in this release is the AI Assistant, designed to make working with Flare both faster and easier. The assistant comes with a Chat and a Code model . The Chat model can answer questions about general Flare usage as well as the underlying science methods. Whether you need a quick reminder on how to set up an experiment, or an explanation of the theory behind a calculation, the Chat model provides guidance directly within your working environment.
The Code model works as a copilot, i.e. a powerful assistant for anyone using the Flare Python API. It can help you write and customize scripts, turning ideas into scripts quickly. Once created, scripts can be run directly in Flare, making it simple to extend functionality or automate routine tasks.
Together, the AI Assistant makes Flare more accessible for new users while empowering advanced users to work faster and share their workflows.
Protein-Protein Docking with Constraints
The capability to impose constraints in a PPD experiment introduces a significant enhancement to this popular method. By strategically placing constraints, it is now possible to dock two protein chains together while biasing their spatial orientation with respect to each other. The constraints ensure that the relative orientations of the proteins are both physically realistic and compatible with the intended binding geometry. This provides a great method for generating ternary complex starting structures.
The design of targeted protein degraders is an example where constrained Protein-Protein Docking is a critical step. Once the proteins are docked under these controlled conditions, the system can serve as the initial scaffold for further studies. With the proteins properly oriented and the binding site left accessible, ligands or linker molecules can be introduced in subsequent modeling steps.
Detached calculations
The new detached calculations capability in Flare V11 connected to Cresset Engine Broker (CEB) V3.5 introduces an efficient way to manage computationally intensive tasks. So far, long-running calculations submitted from the GUI required keeping the Flare GUI active, which limited flexibility. With detached calculations, once a job is launched, Flare transfers all required input data to the CEB and can then safely disconnect. This means users are free to close Flare without interrupting their workflows.
Behind the scenes, the CEB automatically spawns a dedicated background process to handle the calculation independently. Progress and error information are tracked and continually updated in a dedicated Flare window. When the computation is complete, the results are stored in the backend and become immediately available for download and integration into the active project.
This new architecture provides several advantages: it reduces the risk of losing work due to accidental closures or system reboots, lowers local resource consumption by decoupling the GUI from the calculation, and enables more flexible use of compute resources for large or time-consuming tasks.
Prevalence score
Flare V11 brings Prevalence Score, a new method designed to help Flare users quickly assess how common the fragments of the molecules of interest are. This can be linked to the molecule’s synthesizability. Based on fragment frequency data from curated Spark databases (ChEMBL, eMolecules Commercial screening and eMolecules Reagents) the method provides a simple, quantitative estimate of synthetic accessibility.
The Prevalence Score works by fragmenting a molecule and considering how frequently similar fragments appear in the aforementioned databases and thus, in synthetically accessible compounds. Common, easily available fragments are scored favorably, with scores closer to 1, while rare or structurally challenging ones get scores closer to 0. The scores of the fragments can be visualized in Flare. The minimum score is reported as the Prevalence Score of the whole molecule. This is a value ranging from 0 (very low prevalence/very complex) to 1 (high prevalence/very simple), and it is captured in the Prevalence Score column in the Ligands Table.
To make interpretation even easier, Flare also provides a ‘Color by Prevalence score’ option that highlights the contribution of each atom/fragment to the Prevalence Score. By coloring ligands from green to red, users can instantly spot the structural regions that are rare and may hinder synthesis. This interactive view makes it straightforward to identify problematic parts of a molecule, guiding design decisions towards compounds that are practical to make.
Enhanced performance and Nested roles
Flare V11 has been completely rewritten to handle significantly larger projects. Researchers working with tens of thousands of ligands and hundreds of proteins will now find Flare significantly more responsive and reliable. Thanks to a more efficient architecture, memory usage has also been significantly reduced, making it easier to work with large datasets without hitting system limits.
Another highlight of this release is the new ‘nested ligand roles’ feature. This introduces a hierarchical structure within the Ligands table, enabling users to assign ‘parent’ and ‘child’ roles to molecules. This is especially useful in cases where a single chemical entity is represented by multiple variants, such as protonation states, tautomers, or docked conformations. Creating a nested role is straightforward: it can be defined when setting up a new role by choosing a parent, or by using the slash syntax in the Python API. This new flexibility allows researchers to better organize their data and streamline analysis across complex chemical relationships.
Flare FEP improvements enable fast calculations with peptides
Flare FEP in V11 comes with several improvements, that aim to improve both speed and usability.
A key enhancement is the ability to calculate OpenFF partial charges for ligands using machine learning through AshGC (NaGL). This new method delivers extremely fast charge assignment without losing accuracy, making it practical to parameterize large and complex molecules such as peptides and smoothly incorporate them into FEP studies. This enhancement also applies to all calculations that include parameterization of small molecules, such as Minimization, Dynamics, and GIST.
FEP graph generation has been enhanced with significant speed gains. This performance boost is especially valuable when working with large molecules and peptide systems, allowing researchers to set up and execute peptide FEP calculations more efficiently than ever before.
To support analysis, a new ‘Overview’ tab (shown in Figure 8) has been added to Flare FEP projects. This tab presents the FEP results into a single view, displaying predicted and experimental ΔGs (where available) as well as structural information for the ligands involved. Users can now quickly assess the quality of their results and easily export and present their results to the team.
For Absolute FEP calculations, the parameters for setting Boresch restraints and the algorithm for determining constraint strengths have been carefully adjusted based on results of further extensive scientific validation. Restraint definitions now more accurately reflect typical bonded and non-bonded interactions between proteins and ligands. Furthermore, when running only Absolute FEP calculations, an offset relative to experimental values is now calculated and applied directly in the Activity plot, providing a clearer representation of predictive performance.
New features and enhancements in Dynamics
Several key improvements have been introduced to molecular dynamics calculations, extending both the scope of supported systems and the tools available for trajectory analysis.
The AMBER force field has been upgraded to support the ZAFF zinc force field, enabling more accurate simulations of metalloproteins and zinc-containing complexes. Trajectories can also be exported in XTC format (see Figure 9), providing a new widely compatible and storage-efficient option.
Workflow performance has also been refined. The automatic updating of RMSD plots during a Dynamics simulation has been disabled, avoiding unnecessary overhead. Users can now manually refresh the plot when required, ensuring greater control over computational resources.
Analysis and visualization options have been improved as well. Principal Component Analysis (PCA) can now be performed on a user-defined subset of frames rather than being restricted to the full trajectory, giving researchers finer control over the scope of conformational sampling they wish to explore. A speed playback control has also been added to the trajectory movie viewer, making it easier to inspect motions at different timescales and focus on key structural events.
Other highlights of Flare V11
Each Flare release is shaped by two priorities: bringing new scientific developments to the platform and making everyday use smoother and more intuitive. Many of these improvements are inspired by feedback from our community. Below are some more of the standout new features and enhancements arriving in Flare V11:
- Support of 5-letter residues names for all Flare calculations
- New templates for chair and twist-boat 6-member ring conformations in the Editing menu
- Flickering waters are enabled for ‘Score only’ calculations in Flare, as well as for Spark Docking calculations
- New library enumeration reactions, with geminal diols to ketones/ aldehydes and 7-oxabicycloheptane aromatization added to the list of existing reactions
- New option to export the Activity Miner Top Pairs table as SDF
Accelerate your research with Flare V11
Flare V11 introduces powerful new scientific capabilities, faster workflows, and enhanced usability, giving you the tools to explore complex systems with greater efficiency and accuracy.
Get in touch with us today to request an evaluation and gain access to Flare’s latest features. Our expert team will guide you through installation and setup, while our extensive library of tutorials, from essential workflows to advanced methods, ensures a smooth start. With Flare V11, you’ll be equipped to move faster, dig deeper, and design the molecules that matter most.
