From Conversation to Computation: PlayMolecule AI Now Executes Computational Workflows

Recently, we introduced "Speak to a Protein". a new capability in PlayMolecule AI that transformed how researchers interact with molecular data. By using natural language, you could ask complex questions to retrieve, synthesize, and visualize information from databases like the PDB, UniProt, and ChEMBL. This AI co-scientist could identify residues, align structures, and analyze existing data in real time.

That was the first step: enabling analysis. We are now announcing the logical extension: enabling execution.

PlayMolecule AI can now run a comprehensive suite of validated computational chemistry tools. It has evolved from an analytical partner that can find information to a research assistant that can generate new data by running experiments for you, all from the same chat interface.

From Analysis to Execution: An Integrated Toolbox

Previously, you could ask, "Show me the binding site for 4GIH." Now, you can issue a command for an entire workflow: "Find the binding sites in 4GIH, prepare my ligands from this file, dock them, and then predict their binding affinity."

We have integrated Acellera's computational suite directly into the AI agent, giving it a toolbox of capabilities to execute on your behalf. This allows PlayMolecule AI to run complex tasks that generate new predictive data.

The agent's capabilities available in the platform include:

• Full System Preparation: run ProteinPrepare for protonation and structure fixing, SystemBuilder to solvate and add ions, and MembraneBuilder to embed proteins in lipid bilayers, preparing systems for simulation.
• Ligand Design & Docking: Beyond simple visualization, use AcePrep to prepare ligand libraries, Parameterize to assign force field parameters (e.g., GAFF2, OpenFF-Sage), and Generative design tools to suggest new analogs. It can then use DeepSite to identify pockets and AceDock or Boltz-2 to predict ligand poses.
• Affinity Prediction & ML: generate new predictions using Kdeep (our structure-based ML model for affinity) or help you train custom models on your own datasets with KdeepTrainer. It can also generate detailed 2D interaction maps using PlexView.
• Molecular Dynamics (MD): execute MD runs. This includes a SimpleRun for quick equilibrations or full-system setups using SystemBuilder as a preliminary step.
• Data & Cheminformatics: process user-uploaded SDF or CSV files, perform cheminformatics operations using moleculekit or RDKit, and fetch data from UniProt or ChEMBL to initiate a new workflow.

Automating End-to-End Computational Workflows

The primary advantage is PlayMolecule AI's ability to chain these tools into coherent, multi-step workflows. This removes the need for users to write complex scripts, manage intermediate files, or manually transition between different software packages. You simply state your scientific objective.

This enables practical, end-to-end scenarios:

• Target-to-Hit Workflow: A user can state, "Prepare protein 4GIH at pH 7.4, detect its binding pockets, prepare my list of ligands from this SDF file and dock them." PlayMolecule AI will propose the exact steps, run the pipeline, and return the final results.
• Membrane Protein MD Setup: A more complex request, such as, "Build a 100x100 Å POPC membrane, embed my GPCR system, solvate and ionize the system to 0.15M NaCl, and run a 10ns NPT equilibration," can be handled as a single instruction.

Advanced Capabilities for Enterprise Teams

The capabilities described above provide a comprehensive toolkit for a wide range of discovery tasks. For research teams requiring more rigorous, physics-based, and high-throughput simulation, our internal and enterprise versions of PlayMolecule AI integrate an even more powerful set of tools.

The key extensions focus on higher-accuracy simulation and advanced molecular analyses:

• Physics-Based Binding Free Energies: The most significant addition is the ability to run high-accuracy Relative Binding Free Energy (RBFE) calculations. This allows for the precise, physics-based ranking of congeneric ligand series, moving beyond ML-based scoring to rigorous alchemical free energy methods.
• Advanced MD and Analysis: The enterprise suite includes more sophisticated MD protocols, such as mixed-solvent or cosolvent simulations to identify hotspots and cryptic pockets. It also features advanced post-processing, including MSM-driven clustering and the generation of pharmacophore density cubes from trajectories.
• Deeper Pharmacophore and Shape Modeling: This version adds advanced tools for pharmacophore map computation, pose scoring against pharmacophore models, and 3D shape/pharmacophore-based alignment and similarity searches.
• Enhanced Cheminformatics: The toolkit is expanded with more advanced methods like substructure/moiety-based environment mining across the PDB and CLIP-like molecular embedding for nearest neighbor analysis.
• Tighter Integration: The platform includes interactive Python access directly within the viewer for rapid, custom analysis and more robust handling of asynchronous background jobs for long-running simulations.

The Evolving Research Assistant

With "Speak to a Protein," our goal was to lower the barrier for structural analysis. By integrating a full computational toolbox, we aim to lower the barrier for advanced computational experiments.

PlayMolecule AI has evolved from a tool that retrieves information to a platform that can generate and test new hypotheses. This approach makes advanced simulation and machine learning models more accessible, while our enterprise version provides the full power of rigorous, physics-based simulation for the most demanding research programs.

To learn more visit: playmolecule.ai

Try the free version today: open.playmolecule.org

Book your Strategic Consultation & Platform Demo

‍