computational-chemistry-and-biology

name	computational-chemistry-and-biology
description	Run computational chemistry and structural biology calculations on the Rowan platform through Rowan MCP tools or the `rowan` Python package. Use when the user wants to run Rowan workflows, or mentions Rowan, computational chemistry, structural biology, or molecular or protein simulation.

Computational Chemistry and Biology with Rowan

Run simulations on the Rowan platform through Rowan MCP tools or the rowan Python package. This skill provides shared scientific strategy and workflow guidance for both interfaces.

Interfaces are for syntax; skill is for strategy. Use this skill and its workflow references to choose scientifically appropriate methods and settings. Trust discover_workflow or the installed package when an exact parameter differs from a reference.

Choose an interface

Both interfaces are first-class. Use the shared scientific guidance and workflow references below with either one.

• MCP-connected chat: when Rowan tools are available, read MCP execution before calling them. Do not install Python or request an API key; authentication belongs to the MCP connection. Use discover_workflow as the exact parameter contract.
• Python code, scripts, or notebooks: when the user requests Python or no Rowan MCP tools are available, read Python SDK before writing or running code. Trust the installed package when its exact API differs from an example.

For project or folder management, also read folders and projects. For callbacks, read webhooks.

Molecule inputs

Structure-based workflows take a rowan.Molecule, not a bare SMILES string. Build one from a SMILES with rowan.Molecule.from_smiles(...) (RDKit embeds a 3D structure) or from coordinates with rowan.Molecule.from_xyz(...) / from_xyz_file(...). A plain SMILES string is accepted only by the SMILES-based (2D) workflows. Each reference file states which case its workflow is:

• SMILES (2D): predicts from the graph, ignores 3D, so a plain SMILES string works (ADMET, solubility, macropKa, the SMILES pKa/permeability methods).
• 3D from a SMILES is fine: pass rowan.Molecule.from_smiles(...); the embedded structure is built and optimized as needed (basic calculation, NMR, descriptors, redox, BDE, conformer search, and most QM workflows).
• Real coordinates needed: pass a structure with meaningful geometry via rowan.Molecule.from_xyz(...), a prior result's .molecule, or a docked/crystal pose (electronic properties, interaction-energy decomposition, strain, IRC, double-ended TS, 3D pKa methods).

For higher rigor, start a 3D workflow from the conformer search workflow's lowest-energy conformer.

Protein inputs

After loading a protein with rowan.create_protein_from_pdb_id or rowan.upload_protein, prepare it before any docking, MD, or FEP workflow. The default protein.prepare() fixes nonstandard residues, adds missing atoms, and adds hydrogens. See reference/protein_prep.md for prepare and validate_protein_forcefield, options, and the common validation-failure fix.

Workflows

Pick the workflow matching the task by reading the descriptions, then open its reference file (linked at the end of each entry) for recommended settings and an example. Python's universal submit arguments live in Python SDK and aren't repeated per workflow; MCP parameters come from discover_workflow.

Not every account can run every workflow. MCP users must consult the mcp_supported_workflows field returned by account_status; Python users must consult rowan.whoami().enabled_workflows (a few workflows, such as FEP, are gated by plan).

• ADMET: predict ADME-Tox properties (absorption, distribution, metabolism, excretion, toxicity) with an ML model, for fast early developability, PK, and tox triage. See reference/admet.md.
• Analogue docking: pose analogues of an already-bound reference ligand into consistent, analogous poses, to align and prepare a congeneric series, for example ahead of an RBFE screen. See reference/analogue_docking.md.
• Basic calculation: compute core QM results for a single molecular or periodic (crystal/surface) structure, covering single-point energies, geometry and transition-state optimizations (including in solvent), and frequencies and thermochemistry, with DFT, NNP, or semiempirical methods. See reference/basic_calculation.md.
• Batch docking: fast docking and scoring of many protein-ligand poses (AutoDock Vina or QVina2), keeping only the final scores rather than poses, for high-throughput virtual screening of large libraries. See reference/batch_docking.md.
• Binding affinity: SQM-based binding affinity for one or more protein–ligand poses — useful for post-docking triage before RBFE, or when ligand topologies are too dissimilar for FEP. See reference/binding_affinity.md.
• BDE: homolytic bond-dissociation energies for selected bonds, to rank radical stability, find the weakest bonds, or flag likely metabolic or degradation soft spots. See reference/bde.md.
• Conformer search: explore conformational space with fast methods, then rank the generated conformers with a more accurate final method. See reference/conformer_search.md.
• Covalent inhibitor scan: scan the distance between a reactive protein atom and a reactive ligand atom (covalent to non-covalent, or the reverse), to map out the reaction energetics of covalent bond formation. See reference/covalent_inhibitor_scan.md.
• Descriptors: compute electronic and cheminformatic molecular descriptors, to featurize molecules for cheminformatics and machine learning. See reference/descriptors.md.
• Docking: dock a ligand into a protein site and return the pose, refined with a constrained NNP optimization, plus optional conformer search (for a per-pose strain estimate) and pre-docking geometry optimization, for accurate single-ligand pose prediction. Standard and covalent docking are both supported. See reference/docking.md.
• Double-ended TS search: find the transition state connecting a reactant and a product structure, for when you have both endpoints rather than a single TS guess. See reference/double_ended_ts_search.md.
• Electronic properties: compute molecular orbitals, electron density, electrostatic potential, atom-centered charges, bond orders, and multipole moments for a given structure (does not optimize geometry), to analyze electronic structure, reactivity, and charge distribution. See reference/electronic_properties.md.
• Fukui: atom-centered Fukui indices (per-site reactivity toward nucleophiles, electrophiles, and radicals) plus the global electrophilicity index, to predict which sites react and how electrophilic a molecule is overall. See reference/fukui.md.
• Hydrogen-bond acceptor/donor strength: predict hydrogen-bond acceptor (pKBHX) and donor (pKα) strengths using neural network potentials and r2SCAN-3c DFT. See reference/hydrogen_bond_donor_acceptor_strength.md.
• Interaction energy decomposition: break the non-covalent interaction energy between two molecules into SAPT0 components (electrostatics, exchange, induction, dispersion), to quantify how much each contributes. See reference/interaction_energy_decomposition.md.
• Ion mobility: predict a molecule's rotationally averaged collision cross-section (CCS) in nitrogen via explicit trajectory-based scattering, accounting for conformers and protonation states. See reference/ion_mobility.md.
• IRC: trace the steepest-descent reaction path in both directions from an optimized transition state to the reactant and product it connects. See reference/irc.md.
• macropKa: enumerate a molecule's protonation microstates and conformers to predict its macroscopic pKa, microstate populations, and pH-dependent properties (isoelectric point, logD, aqueous solubility, blood-brain-barrier permeability). See reference/macropka.md.
• Membrane permeability: predict small-molecule permeability, either experimental Caco-2 apparent permeability via a graph neural network, or intrinsic permeability coefficients for five membranes (plasma, blood-brain barrier, Caco-2, black lipid membrane, PAMPA) via a physics-based free-energy method. See reference/membrane_permeability.md.
• MSA: generate and format multiple-sequence-alignment data for downstream use with co-folding models. See reference/msa.md.
• Multistage optimization: quickly optimize a structure through successively more accurate levels of theory, yielding an accurate geometry and energy. See reference/multistage_optimization.md.
• NMR: predict a molecule's NMR spectrum. See reference/nmr.md.
• pKa: predict microscopic (per-site) pKa values, with a choice of ML-based and semiempirical methods. See reference/pka.md.
• Pocket detection: detect potential binding pockets on a protein structure. See reference/pocket_detection.md.
• Pose-analysis MD: run MD on a docked protein-ligand complex to test pose stability. See reference/pose_analysis_md.md.
• Protein binder design: generate, filter, and rank protein binders for a given protein or small-molecule target. See reference/protein_binder_design.md.
• Protein co-folding: predict 3D structures of biomolecules and protein-ligand complexes using AlphaFold 3-style models (Boltz-2, Chai-1r, Boltz-1). See reference/protein_cofolding.md.
• Protein MD: run a molecular dynamics simulation on a protein. See reference/protein_md.md.
• RBFE graph: take a list of ligands and generate the graph used by the relative binding free energy perturbation (FEP) workflow. Its edges are the ligand pairs FEP transforms between. See reference/rbfe_graph.md.
• Redox potential: predict a molecule's reduction or oxidation potential. See reference/redox_potential.md.
• Relative binding free energy perturbation: predict relative binding affinities across a congeneric ligand series using free energy perturbation (FEP), to rank analogs and prioritize which to synthesize. See reference/relative_binding_free_energy_perturbation.md.
• Scan: compute how a system's energy changes along a scan coordinate (bond, angle, or dihedral). See reference/scan.md.
• Solubility: predict temperature-dependent solubility in a variety of solvents, plus room-temperature aqueous solubility from dedicated aqueous-solubility models. See reference/solubility.md.
• Solvent-dependent conformers: generate, optimize, and score conformers across multiple solvents to see how conformational preferences shift between them. See reference/solvent_dependent_conformers.md.
• Spin states: predict a molecule's lowest-energy spin state by optimizing it across different spin states. See reference/spin_states.md.
• Strain: compute the strain energy of a specific pose by comparing it to the global-minimum conformer. See reference/strain.md.
• Tautomer search: enumerate tautomers and rank their relative stability using neural network potentials or semiempirical methods. See reference/tautomer_search.md.