tooluniverse-binder-discovery

What it does

The skill systematically discovers novel small molecule binders for a protein target by combining structure-based and ligand-based approaches across druggability assessment, known ligand mining, similarity expansion, ADMET filtering, and synthesis feasibility. It emphasizes a report-first workflow that creates a binder discovery report file first, then progressively updates it with findings, and outputs separate data files for prioritized candidates and literature references.

How it works

• It follows a multi-stage workflow starting with Phase 0 Tool Verification to check parameters for tools like ChEMBL_get_target_activities.
• Phase 1 Target Validation performs identifier resolution, drudgability assessment, binding-site identification, and structure prediction where needed, including GPCR-specific enrichment and optional antibody landscape checks via Thera-SAbDab.
• Phase 2 Known Ligand Mining gathers bioactivity data from ChEMBL, ligand interactions from GtoPdb, chemical probes from OpenTargets, and SAR/actives analysis, with outputs embedded into the report.
• Phase 3 Structure Analysis retrieves PDB structures, assesses binding sites, and analyzes domain architecture via InterPro data.
• Phase 4 Compound Expansion explores similar structures and performs de novo generation when appropriate.
• Phase 5 ADMET Filtering predicts properties and flags liabilities.
• Phase 6 Candidate Docking & Prioritization docks candidates, scores by docking/ADMET/novelty, assesses synthesis feasibility, and generates a final ranked list.
• Phase 7 Report Synthesis consolidates the results into the final report. The system maintains a strict report-first approach, creating a file named [TARGET]_binder_discovery_report.md with placeholder sections, then filling them progressively.

When to use it

Use when users ask to find small molecules for a target, identify novel binders, perform virtual screening, or need hit-to-lead compound identification.

What it can touch

• Tool names referenced include: ChEMBL_get_target_activities, NVIDIA_API_KEY is required for NVIDIA NIM actions, and tools like NvidiaNIM_alphafold2, NvidiaNIM_esmfold may be invoked for structure prediction. The workflow specifies creating and updating files such as [TARGET]_binder_discovery_report.md, [TARGET]_candidate_compounds.csv, and [TARGET]_bibliography.json as output artifacts.

Caveats

• The workflow relies on environment and tool availability (e.g., NVIDIA NIM and API keys) and requires careful phase transitions to maintain the report-first discipline. No outcomes are guaranteed; results depend on input data and tool responses. The approach includes explicit reporting formats and source citations in the report sections.