laboratoire-de-chemoinformatique/chemspacecopilot/candidate-design-to-gtm

Candidate Design To GTM

Use this skill when the user wants generated small-molecule or peptide candidates prepared for GTM projection.

Procedure

1. Determine whether the task is small-molecule or peptide design.
2. Resolve any seed compound, peptide sequence, or prior candidate-set reference from session memory before generation.
3. For molecules, use mol_design_molecules or mol_generate_analogs, then mol_validate_design_candidates, mol_rank_design_candidates, and mol_register_design_candidates.
4. For peptides, use peptide_design_peptides or peptide_generate_analogs, then peptide_validate_design_candidates, peptide_rank_design_candidates, and candidate registration/materialization through the session tools available for peptide candidate sets.
5. Choose the engine: engine="autoencoder" (molecules) or engine="wae" (peptides) run in-process. engine="llm" is delegated to the LLM — under the default MCP llm_policy="external" the design tool returns a needs_external_llm task you (the outer agent) complete via llm_get_task / llm_submit_task_result; the Agno team uses its own model inline. Only llm_policy="disabled" rejects it.
6. Materialize the registered candidate set with session_materialize_candidate_set_dataset.
7. Use gtm_project_data only when a suitable GTM model is available or the user requested projection.

Expected Outputs

• Registered candidate set.
• Candidate dataset CSV.
• Optional GTM projection artifact.
• Validation and ranking summary.