tooluniverse-admet-prediction

What this skill does

# ADMET Prediction & Drug Candidate Profiling

**ADMET reasoning**: a drug fails if it can't be absorbed, distributes to wrong tissues, isn't metabolized safely, or isn't excreted. Evaluate each property independently — good absorption doesn't compensate for liver toxicity. The ADME properties determine whether a compound reaches its target at therapeutic concentrations; toxicity determines whether it's safe to do so. Prioritize experimental data (T2) over computational predictions (T3) — ADMETAI predictions are screening tools, not definitive verdicts. When a FAIL is flagged in any toxicity category (hERG, AMES, DILI), treat it as program-limiting until wet-lab data refutes it.

**LOOK UP DON'T GUESS**: never assume SMILES, CID, or experimental LD50 values — always call PubChem to resolve compound identity before any ADMETAI or PubChemTox call.

Comprehensive pharmacokinetic and toxicity profiling integrating AI-based ADMET predictions, rule-based drug-likeness filters, and experimental benchmarks from curated databases.

## When to Use This Skill

**Triggers**:
- "What are the ADMET properties of [compound]?"
- "Is [drug] likely to cross the blood-brain barrier?"
- "Predict the toxicity of this SMILES: ..."
- "Does [compound] violate Lipinski's rule of five?"
- "Assess the drug-likeness of [molecule]"
- "What are the CYP interactions for [drug]?"
- "Pharmacokinetic profile of [compound]"
- "Is [compound] orally bioavailable?"
- "What is the LD50 / hERG liability of [molecule]?"

**Input**: Drug name (e.g., "ibuprofen") OR SMILES string (e.g., "CC(C)Cc1ccc(cc1)C(C)C(=O)O")

---

## COMPUTE, DON'T DESCRIBE
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.

## KEY PRINCIPLES

1. **Resolve identity first** - Always convert drug name to SMILES before calling ADMETAI tools
2. **ADMETAI tools require `tooluniverse[ml]`** - If import fails, skip to SwissADME/PubChemTox fallbacks
3. **All ADMETAI tools take `smiles: list[str]`** - Always wrap in a list, even for one compound
4. **SwissADME takes `smiles: str`** - Single string, NOT a list (SOAP-style with `operation` param)
5. **PubChemTox tools accept `cid` or `compound_name`** - Use CID when available for reliability
6. **Evidence grading mandatory** - Predictions (T3), experimental data (T2), regulatory (T1)
7. **Scorecard output** - Every analysis must end with a pass/warn/fail scorecard
8. **Explain significance** - State WHY each property matters for drug development

---

## Evidence Grading

| Tier | Label | Source |
|------|-------|--------|
| **T1** | Regulatory/Clinical | FDA labels, ChEMBL max clinical phase |
| **T2** | Experimental | PubChemTox LD50/LC50, in vitro AMES, animal studies |
| **T3** | Computational | ADMETAI predictions, SwissADME calculations |
| **T4** | Annotation | Database cross-references, text-mined |

## Workflow: 5-Phase ADMET Profiling

```
User Query (drug name or SMILES)
|
+-- PHASE 1: Compound Identity Resolution
|   PubChem name->CID->SMILES, or validate input SMILES
|
+-- PHASE 2: Physicochemical & Drug-Likeness
|   ADMETAI physicochemical + SwissADME druglikeness -> Lipinski/Veber
|
+-- PHASE 3: ADME Predictions
|   BBB, bioavailability, CYP interactions, clearance, solubility
|
+-- PHASE 4: Toxicity Assessment
|   ADMETAI tox + PubChemTox experimental + nuclear receptor + stress
|
+-- PHASE 5: Scorecard & Clinical Context
|   ChEMBL max phase, aggregate pass/warn/fail, final recommendation
```

---

### PHASE 1: Compound Identity Resolution

**Goal**: Obtain SMILES, PubChem CID, and basic identifiers for the query compound.

**Steps**:

1. **If input is a drug name**:
   - Call `PubChem_get_CID_by_compound_name(name=<drug_name>)` to get CID
   - Call `PubChem_get_compound_properties_by_CID(cid=<CID>)` to get SMILES and MW
   - Extract `ConnectivitySMILES` from the response (NOT `CanonicalSMILES`)

2. **If input is a SMILES string**:
   - Call `PubChem_get_CID_by_SMILES(smiles=<SMILES>)` to get CID
   - Call `PubChem_get_compound_properties_by_CID(cid=<CID>)` for compound name and MW
   - Use the input SMILES for all subsequent ADMETAI calls

3. **Record**:
   - Compound name, CID, SMILES, molecular formula, molecular weight, IUPAC name
   - If CID lookup fails, proceed with SMILES only (ADMETAI does not need CID)

**Why this matters**: ADMETAI tools require SMILES input. PubChemTox tools work best with CID. Resolving both ensures all downstream tools can be called. PubChem is the authoritative source for SMILES canonicalization.

**Fallback**: If PubChem has no entry, the user must provide SMILES directly. Cannot proceed without SMILES.

---

### PHASE 2: Physicochemical Properties & Drug-Likeness

**Goal**: Evaluate whether the compound has drug-like physicochemical properties.

**Steps**:

1. **ADMETAI physicochemical** (primary):
   ```
   ADMETAI_predict_physicochemical_properties(smiles=["<SMILES>"])
   ```
   Returns: MW, logP, TPSA, HBD, HBA, rotatable bonds

2. **SwissADME drug-likeness** (complementary):
   ```
   SwissADME_check_druglikeness(operation="check_druglikeness", smiles="<SMILES>")
   SwissADME_calculate_adme(operation="calculate_adme", smiles="<SMILES>")
   ```
   Returns: Lipinski, Veber, Ghose, Egan, Muegge rule compliance; PAINS alerts; Brenk alerts

3. **ADMETAI solubility**:
   ```
   ADMETAI_predict_solubility_lipophilicity_hydration(smiles=["<SMILES>"])
   ```
   Returns: Aqueous solubility (LogS), lipophilicity, hydration free energy

**Interpret & Score**:

| Property | Ideal Range | Why It Matters |
|----------|-------------|----------------|
| MW | < 500 Da | Larger molecules have poor membrane permeability (Lipinski) |
| LogP | -0.4 to 5.6 | Too hydrophobic = poor solubility; too hydrophilic = poor permeability |
| HBD | <= 5 | Excess donors reduce membrane crossing (Lipinski) |
| HBA | <= 10 | Excess acceptors reduce membrane crossing (Lipinski) |
| TPSA | < 140 A^2 | High PSA correlates with poor oral absorption |
| Rotatable bonds | <= 10 | Molecular flexibility affects bioavailability (Veber) |
| LogS | > -6 | Below -6 = practically insoluble, formulation challenge |
| PAINS alerts | 0 | Pan-assay interference compounds give false positives in screens |

**Verdict**: PASS if Lipinski <= 1 violation and no PAINS alerts; WARN if 2 violations; FAIL if 3+ violations or PAINS+.

**Fallback**: If ADMETAI import fails (missing `tooluniverse[ml]`), rely on SwissADME alone. SwissADME provides all Lipinski descriptors independently.

---

### PHASE 3: ADME Predictions

**Goal**: Predict absorption, distribution, metabolism, and excretion behavior.

**Steps**:

1. **Blood-brain barrier penetration**:
   ```
   ADMETAI_predict_BBB_penetrance(smiles=["<SMILES>"])
   ```
   - BBB+ = compound can cross; BBB- = cannot
   - Critical for CNS drugs (must cross) and peripherally-acting drugs (should NOT cross to avoid CNS side effects)

2. **Oral bioavailability**:
   ```
   ADMETAI_predict_bioavailability(smiles=["<SMILES>"])
   ```
   - F20% = at least 20% oral bioavailability; F30% = at least 30%
   - Low bioavailability means the drug is extensively metabolized or poorly absorbed
   - F < 20% generally requires non-oral routes (IV, inhaled, topical)

3. **CYP450 interactions**:
   ```
   ADMETAI_predict_CYP_interactions(smiles=["<SMILES>"])
   ```
   - Reports substrate/inhibitor status for CYP1A2, 2C9, 2C19, 2D6, 3A4
   - **Why CYP matters**: ~75% of drugs are metabolized by CYP enzymes. Inhibiting CYP3A4 (which metabolizes ~50% of drugs) causes dangerous drug-drug interactions (DDIs). CYP2D6 polymorphisms affect ~25% of drugs -- poor metabolizers accumulate toxic levels
   - Substrate of CYP2D6 = pharmacogenomic risk (poor/ultra-rapid metabolizers)
   - Inhibitor of CYP3A4 = high DDI risk