Help us improve
Share bugs, ideas, or general feedback.
From tooluniverse
Scores drug target druggability, selectivity, safety, ADMET, and structural tractability, producing a composite score (0-100) and GO/NO-GO recommendation for target prioritization and de-selection.
npx claudepluginhub mims-harvard/tooluniverse --plugin tooluniverseHow this skill is triggered — by the user, by Claude, or both
Slash command
/tooluniverse:tooluniverse-drug-target-validationThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Validate drug target hypotheses using multi-dimensional computational evidence before committing to wet-lab work. Produces a quantitative Target Validation Score (0-100) with priority tier classification and GO/NO-GO recommendation.
Gathers comprehensive drug-target intelligence from 9 parallel research paths: tissue expression, pathways, protein interactions, variants, druggability, with citations. For target characterization and reports.
Scores drug discovery targets with GO/NO-GO decisions using evidence from Open Targets, ChEMBL, PDB, and ClinicalTrials.gov. For prioritizing targets in campaigns.
Queries Open Targets Platform GraphQL API for target-disease associations, drug target discovery, tractability/safety data, genetics/omics evidence, and known drugs.
Share bugs, ideas, or general feedback.
Validate drug target hypotheses using multi-dimensional computational evidence before committing to wet-lab work. Produces a quantitative Target Validation Score (0-100) with priority tier classification and GO/NO-GO recommendation.
A valid drug target must pass 4 gates in order. Failing an early gate makes later gates irrelevant:
Do not proceed to Phase 3 (Chemical Matter) before completing Phase 1 (Disease Association). Gate 1 failures should prompt a NO-GO or pivot recommendation.
LOOK UP DON'T GUESS: Never assume a target is druggable based on its protein family alone, never assume expression is low in a tissue without checking GTEx or HPA, never assume no competitors without searching ClinicalTrials.gov.
RUN THE ML MODELS, DON'T SKIP THEM: When deep-learning predictors are available (ADMET-AI, ESMFold, AlphaFold, DoGSite, DynaMut2, DeepGO), run them even when database lookups or experimental data already cover the same property. The ML predictions provide an orthogonal, mechanistically-grounded estimate that's a first-class output of this skill — not a fallback. A target-validation report missing ML predictions is incomplete regardless of how much database evidence is present.
When analysis requires computation (statistics, data processing, scoring, enrichment), write and run Python code via Bash. Don't describe what you would do — execute it and report actual results. Use ToolUniverse tools to retrieve data, then Python (pandas, scipy, statsmodels, matplotlib) to analyze it.
Apply when users ask about:
Not for (use other skills): general target biology (tooluniverse-target-research), drug compound profiling (tooluniverse-drug-research), variant interpretation (tooluniverse-variant-interpretation), disease research (tooluniverse-disease-research).
| Parameter | Required | Description | Example |
|---|---|---|---|
| target | Yes | Gene symbol, protein name, or UniProt ID | EGFR, P00533 |
| disease | No | Disease/indication for context | Non-small cell lung cancer |
| modality | No | Preferred therapeutic modality | small molecule, antibody, PROTAC |
Total: 0-100 points across 5 dimensions (details in SCORING_CRITERIA.md):
| Dimension | Max | Sub-dimensions |
|---|---|---|
| Disease Association | 30 | Genetic (10) + Literature (10) + Pathway (10) |
| Druggability | 25 | Structure (10) + Chemical matter (10) + Target class (5) |
| Safety Profile | 20 | Expression (5) + Genetic validation (10) + ADRs (5) |
| Clinical Precedent | 15 | Based on highest clinical stage achieved |
| Validation Evidence | 10 | Functional studies (5) + Disease models (5) |
Priority Tiers: 80-100 = Tier 1 (GO) | 60-79 = Tier 2 (CONDITIONAL GO) | 40-59 = Tier 3 (CAUTION) | 0-39 = Tier 4 (NO-GO)
Evidence Grades: T1 (clinical proof) > T2 (functional studies) > T3 (associations) > T4 (predictions)
Resolve target to ALL identifiers before any analysis.
Steps:
MyGene_query_genes - Get initial IDs (Ensembl, UniProt, Entrez)ensembl_lookup_gene - Get versioned Ensembl ID (species="homo_sapiens" REQUIRED)ensembl_get_xrefs - Cross-references (HGNC, etc.)OpenTargets_get_target_id_description_by_name - Verify OT targetChEMBL_search_targets - Get ChEMBL target IDUniProt_get_function_by_accession - Function summary (returns list of strings)UniProt_get_alternative_names_by_accession - Collision detectionOutput: Table of verified identifiers (Gene Symbol, Ensembl, UniProt, Entrez, ChEMBL, HGNC) plus protein function and target class.
Quantify target-disease association from genetic, literature, and pathway evidence.
Key tools:
OpenTargets_get_diseases_phenotypes_by_target_ensembl - Disease associationsOpenTargets_target_disease_evidence - Detailed evidence (needs efoId + ensemblId)OpenTargets_get_evidence_by_datasource - Evidence by data sourcegwas_get_snps_for_gene / gwas_search_studies - GWAS evidencegnomad_get_gene_constraints - Genetic constraint (pLI, LOEUF)PubMed_search_articles - Literature (returns plain list of dicts)OpenTargets_get_publications_by_target_ensemblID - OT publications (uses entityId)Assess whether the target is amenable to therapeutic intervention.
Key tools:
OpenTargets_get_target_tractability_by_ensemblID - Tractability (SM, AB, PR, OC)OpenTargets_get_target_classes_by_ensemblID - Target classificationPharos_get_target - TDL: Tclin > Tchem > Tbio > TdarkDGIdb_get_gene_druggability - Druggability categoriesalphafold_get_prediction (param: qualifier) / alphafold_get_summaryProteinsPlus_predict_binding_sites - Pocket detectionOpenTargets_get_chemical_probes_by_target_ensemblID - Chemical probesOpenTargets_get_target_enabling_packages_by_ensemblID - TEPsTCDB_get_transporter - For SLC/ABC transporter targets: TC classification, family, PDB structures (param: uniprot_accession)TCDB_search_by_substrate - Find transporters by substrate (param: substrate_name)Identify existing chemical starting points for target validation.
Key tools:
ChEMBL_search_targets + ChEMBL_get_target_activities - Bioactivity data (note: target_chembl_id__exact with double underscore)BindingDB_get_ligands_by_uniprot - Binding data (affinity in nM)PubChem_search_assays_by_target_gene + PubChem_get_assay_active_compounds - HTS dataOpenTargets_get_associated_drugs_by_target_ensemblID - Known drugs (size REQUIRED)ChEMBL_search_mechanisms - Drug mechanismsDGIdb_get_gene_info - Drug-gene interactionsFor each lead / approved compound identified above, run all ten ADMET-AI Chemprop-GNN endpoints. This is a required deliverable of the skill, not optional:
| Endpoint | Tool |
|---|---|
| Physicochemical (MW, logP, HBA/HBD, TPSA) | ADMETAI_predict_physicochemical_properties |
| Toxicity (AMES, DILI, LD50, carcinogens, skin sensitizers, ClinTox) | ADMETAI_predict_toxicity |
| BBB penetrance | ADMETAI_predict_BBB_penetrance |
| CYP interactions (1A2, 2C9, 2C19, 2D6, 3A4) | ADMETAI_predict_CYP_interactions |
| Bioavailability (HIA, PAMPA, Caco-2, F20/F30) | ADMETAI_predict_bioavailability |
| Clearance & distribution (hepatocyte, microsome, VDss, PPB) | ADMETAI_predict_clearance_distribution |
| Nuclear receptor activity (NR-AR, NR-AhR, NR-Aromatase, NR-ER, NR-PPAR-γ) | ADMETAI_predict_nuclear_receptor_activity |
| Stress response (SR-ARE, SR-ATAD5, SR-HSE, SR-MMP, SR-p53) | ADMETAI_predict_stress_response |
| Solubility, lipophilicity, hydration | ADMETAI_predict_solubility_lipophilicity_hydration |
| Metabolism | ADMETAI_predict_metabolism (if available) |
Required output — ADMET head-to-head table: when two or more candidate drugs exist (approved or late-stage), produce a side-by-side comparison table with every endpoint in the same row and a "Winner" column flagging which drug is safer. This table is the primary visual of the report and must not be abbreviated or summarized into prose.
ADMET-AI fallback (IMPORTANT): If MCP calls to ADMETAI_predict_* fail, return empty, or timeout, run them via Bash + Python SDK instead:
from tooluniverse import ToolUniverse
tu = ToolUniverse()
tu.load_tools()
for endpoint in ['physicochemical_properties','toxicity','BBB_penetrance','CYP_interactions',
'bioavailability','clearance_distribution','nuclear_receptor_activity',
'stress_response','solubility_lipophilicity_hydration']:
r = tu.run_one_function({'name': f'ADMETAI_predict_{endpoint}',
'arguments': {'smiles_list': [SMILES_DRUG_A, SMILES_DRUG_B]}})
print(f'{endpoint}: {r}')
This SDK path bypasses the CLI subprocess and avoids segfault issues with torch. Always try MCP first; use this fallback if MCP returns no data.
Assess clinical validation from approved drugs and clinical trials.
Key tools:
FDA_get_mechanism_of_action_by_drug_name / FDA_get_indications_by_drug_namedrugbank_get_targets_by_drug_name_or_drugbank_id (ALL params required: query, case_sensitive, exact_match, limit)search_clinical_trials (query_term REQUIRED)OpenTargets_get_drug_warnings_by_chemblId / OpenTargets_get_drug_adverse_events_by_chemblIdIdentify safety risks from expression, genetics, and known adverse events.
Key tools:
OpenTargets_get_target_safety_profile_by_ensemblID - Safety liabilitiesGTEx_get_median_gene_expression - Tissue expression (operation="median" REQUIRED)HPA_search_genes_by_query / HPA_get_comprehensive_gene_details_by_ensembl_idOpenTargets_get_biological_mouse_models_by_ensemblID - KO phenotypesFDA_get_adverse_reactions_by_drug_name / FDA_get_boxed_warning_info_by_drug_nameOpenTargets_get_target_homologues_by_ensemblID - Paralog risksCritical tissues to check: heart, liver, kidney, brain, bone marrow.
Understand the target's role in biological networks and disease pathways.
Key tools:
Reactome_map_uniprot_to_pathways (param: id, NOT uniprot_id)STRING_get_protein_interactions (param: protein_ids as array, species=9606)intact_get_interactions - Experimental PPIOpenTargets_get_target_gene_ontology_by_ensemblID - GO termsSTRING_functional_enrichment - Enrichment analysisAssess: pathway redundancy, compensation risk, feedback loops.
Assess existing functional validation data.
Key tools:
DepMap_get_gene_dependencies - Essentiality (score < -0.5 = essential)PubMed_search_articles - Search for CRISPR/siRNA/knockout studiesCTD_get_gene_diseases - Gene-disease associationsLeverage structural biology for druggability and mechanism understanding. ALWAYS run both the deep-learning predictors (ESMFold, DoGSite) AND retrieve experimental structures, even when high-resolution PDB entries already exist. The ML models give an independent pLDDT/druggability score that is a required output of this phase.
Required tool calls (every run):
ESMFold_predict_structure — Meta ESM-2 language-model structure prediction from the UniProt sequence. Report: model pLDDT, worst-residue confidence, RMSD vs. reference PDB if available.alphafold_get_prediction / alphafold_get_summary — DeepMind AlphaFold model + per-residue pLDDT.ProteinsPlus_predict_binding_sites — DoGSite deep-learning pocket scoring. Report: top 3 pockets with volume, druggability score, residue composition.Supporting tools:
UniProt_get_entry_by_accession - Extract PDB cross-referencesget_protein_metadata_by_pdb_id / pdbe_get_entry_summary / pdbe_get_entry_qualityInterPro_get_protein_domains / InterPro_get_domain_details - Domain architectureComprehensive collision-aware literature analysis.
Steps:
"{gene_symbol}"[Title] in PubMed; if >20% off-topic, add filters (AND protein OR gene OR receptor)review[pt] filter in PubMedopenalex_search_works for impact dataEuropePMC_search_articlesSynthesize all phases into actionable output:
| Model | Architecture | Contributed |
|---|---|---|
| AlphaFold | DeepMind iterative SE(3)-equivariant Transformer | Full-length 3D model; per-residue pLDDT 91.5 |
| ESMFold | Meta ESM-2 protein language model | Sequence→structure baseline; confidence vs. AlphaFold |
| DoGSite3 | CNN pocket scorer (ProteinsPlus) | Top-3 druggable pockets with volume and drug-score |
| ADMET-AI | Chemprop GNN ensemble (TDC) | 10 endpoints for sotorasib / adagrasib (table above) |
| DynaMut2 | Graph-based mutation stability predictor | ΔΔG for G12C vs. WT |
| DeepGO | Hierarchical GO-term classifier | Molecular-function predictions |
Only list models actually called during the run. This section makes the ML content first-class for a scientific or investor audience.
Create file: [TARGET]_[DISEASE]_validation_report.md
Use the full template from REPORT_TEMPLATE.md. Key sections:
Complete the Completeness Checklist (in REPORT_TEMPLATE.md) before finalizing to verify all phases were covered, all scores justified, and negative results documented.