skills.rest
matt-grain
matt-grainCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

tooluniverse-binder-discovery

Discover small molecule binders for protein targets using ToolUniverse databases and NVIDIA NIM docking.

Updated Mar 13, 2026
Search Tags:#virtual screening#drug discovery#molecular docking#chembl#binder discovery#nvidia nim#admet prediction
One-click install
npx skills add https://github.com/matt-grain/pharma-catalyst --skill tooluniverse-binder-discovery-matt-grain
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: tooluniverse-binder-discovery
Source: https://github.com/matt-grain/pharma-catalyst/tree/main/.claude/skills/tooluniverse-binder-discovery
Command: npx skills add https://github.com/matt-grain/pharma-catalyst --skill tooluniverse-binder-discovery-matt-grain

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Finding novel small molecule binders for a protein target requires querying dozens of databases, mining bioactivity data, filtering by ADMET properties, and validating with docking — a fragmented, error-prone process. This Skill orchestrates that entire workflow into a single structured pipeline that produces a prioritized, evidence-graded candidate report. ## Core Features & Use Cases - End-to-end discovery workflow: Seven phases covering target validation, known ligand mining (ChEMBL, BindingDB, PubChem), structure analysis (PDB, EMDB, AlphaFold), compound expansion, ADMET filtering, and candidate prioritization. - NVIDIA NIM integration: Structure prediction (AlphaFold2, ESMFold), docking (DiffDock, Boltz2), and de novo molecule generation (GenMol, MolMIM) with fallback chains when the API key is unavailable. - Evidence-graded output: Candidates ranked on a T0–T5 evidence tier scale with multi-factor scoring (docking 40%, ADMET 30%, similarity 20%, novelty 10%), delivered as a structured report and CSV. - Use Case: Ask for novel binders for EGFR and receive a report with druggability assessment, top 10 known actives with IC50 values, 20 ranked candidate compounds with SMILES and ADMET scores, plus an experimental validation plan. ## Quick Start Find novel small molecule binders for the CDK4 kinase and generate a prioritized candidate report with docking scores and ADMET profiles.

Frequently Asked Questions about tooluniverse-binder-discovery

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I find novel small molecule binders for a protein target?▼

Provide a target name and the workflow resolves identifiers via UniProt and ChEMBL, mines known ligands, expands candidates by similarity and substructure search, filters by ADMET properties, and ranks the top 20 compounds with evidence tiers in a structured report.

What databases are used for known ligand mining?▼

The workflow queries ChEMBL bioactivity data as the primary source, with BindingDB for direct Ki/Kd affinities, GtoPdb for pharmacology interactions, and PubChem BioAssay for HTS screening hits. Fallback chains handle missing data automatically.

Does the docking workflow require an NVIDIA API key?▼

NVIDIA NIM tools (AlphaFold2, DiffDock, Boltz2, GenMol, MolMIM) require the NVIDIA_API_KEY environment variable. Without it, the workflow falls back to AlphaFold DB structures and similarity-based scoring instead of docking.

How are candidate compounds prioritized and scored?▼

Candidates are scored on docking confidence (40%), ADMET profile (30%), similarity to known actives (20%), and novelty (10%). Each candidate receives an evidence tier from T0 (beats reference docking score) to T5 (generated molecule without docking).

Why do ToolUniverse calls sometimes return empty results?▼

Empty results usually come from wrong parameter names, such as ensembl_id instead of ensemblId or chembl_target_id instead of target_chembl_id. The skill includes a parameter correction table and recommends verifying tools with get_tool_info before first use.

What are the limitations for novel or poorly characterized targets?▼

Targets with limited bioactivity data yield lower-confidence candidates, mostly T3–T5 evidence tiers. The workflow compensates by mining related target family ligands, using predicted structures, and documenting all data gaps with alternative strategies.

Related Skills

Explore More Education & Research Skills→
jaechang-hitsjaechang-hits
298

gget-genomic-databases

Query over 20 genomic and biological databases via command-line or Python interface.

Community
Intermediate
biomaps-infrabiomaps-infra

db-bindingdb

Query BindingDB for drug-target binding affinities by UniProt ID or SMILES.

Official
Advanced
learningmatter-mitlearningmatter-mit
144

general-chemical-literature

Retrieve PubMed and patent cross-references for a chemical compound via PubChem CID.

Official
Basic
pradyumnasagarpradyumnasagar
6

free-energy-calculations

Execute alchemical free-energy calculations for ligand binding to proteins.

Community
Advanced