Most AI agents can call tools. The harder part is knowing how those tools should actually be used.
With SciAgentKit, computational drug-discovery workflows are guided by skills written by domain experts. These skills encode tool selection, parameter choices, validation checks, and scientific decision rules across ligand preparation, PDB selection, pocket detection, docking, molecular dynamics, and trajectory analysis.
The calculations are performed by established scientific software, while the agent follows expert-defined workflow logic.
The aim is to make advanced drug-discovery workflows more reproducible, consistent, and easier to run through natural language.
Plug-and-play conda/mamba environments per tool would save a ton of setup headaches, especially for folks juggling RDKit, OpenMM, and AutoDock Vina versions across different projects.
@feyza1156028 Thanks, this is a very good point. Dependency conflicts between RDKit, OpenMM and docking tools are one of the biggest practical problems in these workflows.
We are working toward more isolated and reproducible environments, including tool-specific conda/mamba setups and clearer environment validation before a workflow starts.