---
name: virtual-screening-pipeline
description: Perform small molecule virtual screening by combining ligand-based similarity search, molecular docking, and consensus scoring to identify potential drug candidates. Includes ECFP4 fingerprint calculation, Tanimoto similarity, AutoDock Vina, and configurable consensus scoring.
allowed-tools: WebFetch, Bash(python *), Bash(mkdir *), Bash(cp *), Bash(ls *), Bash(jq *), Bash(cd *), Bash(pip *), Bash(rdkit install *)
---

# Small Molecule Virtual Screening Pipeline
## ???: v1.0 | ???: 2026-04-01

## ??????

??ipeline???**??????????????*???????????????$????????????????????n
1. **?????????????(Ligand-Based VS)**: ??????????????????
2. **?????????????(Structure-Based VS)**: ????????????????n3. **?????? (Consensus Scoring)**: ??????????????????????n
**???**: ?????????(PDB) + ??????(SMILES/SDF) + ???????????????
**???**: ???????????????????????????????

---

## ??????: ???????????n
### ???1: ?????? (Target Analysis)

#### 1.1 ?????????

**????????* (???????????:
1. X-ray?????? (?????< 2.5?)
2. ????????? (?????< 3.5?)
3. AlphaFold3 ?????? (pLDDT > 70)
4. AlphaFold2 ?????? (pLDDT > 85)

**?????????????*:
| ??? | ??? | ??? | ?????| ??|
|------|------|------|--------|-----|
| pLDDT (???) | > 90 | 70-90 | 50-70 | < 50 |
| pLDDT (??????) | > 85 | 80-85 | 70-80 | < 70 |
| X-ray?????| < 1.5? | 1.5-2.5? | 2.5-3.5? | > 3.5? |

#### 1.2 ?????????

**?h????????????*:
```
?h???? (Pocket Volume): > 500 ?? ?????n?????? (Hydrophobic Fraction): > 30% ?????n???????????: < 40% ?????n```

---

### ???2: ?????? (Ligand Preparation)

#### 2.1 ?????????

**SMILES??????** (???):
```
ID1 CC(=O)Oc1ccc(cc1)CC(NCCc2ccc(Cl)cc2)=O 432.3
ID2 c1ccc(C(=O)Nc2ccc(N)cc2)cc1 287.3
???: [ID] [SMILES] [?????????]
```

**????????????**:
1. ZINC22 (https://zinc20.docking.org/) - ???????????????
2. ChEMBL (https://www.ebi.ac.uk/chembl/) - ?????????????????
3. Enamine REAL - ????????? (10M+)

#### 2.2 ????????n
```python
from rdkit import Chem
from rdkit.Chem import SaltRemover

def standardize_molecule(smiles):
    # 1. ????????????
    remover = SaltRemover.SaltRemover()
    mol = Chem.MolFromSmiles(smiles)
    mol = remover.StripMol(mol, dontRemoveEverything=True)
    return mol
```

#### 2.3 ???????????n
**Lipinski????????* (?f????):
```
?????(MW):     < 500 Da
?????????(LogP): < 5
????????(HBD):  ??5
????????(HBA):  ??10
????????         ??10
???????? (TPSA):  20-140 ??
```

```python
from rdkit import Chem
from rdkit.Chem import Descriptors, Lipinski

def drug_likeness_filter(mol):
    """????????????????????(???, ?????????)"""
    if mol is None:
        return False, ["Invalid structure"]
    failures = []
    mw = Descriptors.MolWt(mol)
    if mw >= 500:
        failures.append(f"MW={mw:.1f} >= 500")
    logp = Descriptors.MolLogP(mol)
    if logp >= 5:
        failures.append(f"LogP={logp:.2f} >= 5")
    hbd = Lipinski.NumHDonors(mol)
    if hbd > 5:
        failures.append(f"HBD={hbd} > 5")
    hba = Lipinski.NumHAcceptors(mol)
    if hba > 10:
        failures.append(f"HBA={hba} > 10")
    rotatable = Lipinski.NumRotatableBonds(mol)
    if rotatable > 10:
        failures.append(f"RotBonds={rotatable} > 10")
    tpsa = Descriptors.TPSA(mol)
    if tpsa > 140:
        failures.append(f"TPSA={tpsa:.1f} > 140")
    elif tpsa < 20:
        failures.append(f"TPSA={tpsa:.1f} < 20")
    return len(failures) == 0, failures
```

---

### ???3: ?????????????n
#### 3.1 ???????$?

**ECFP4???** (?????):
```python
from rdkit import Chem
from rdkit.Chem import AllChem

def calculate_ecfp4(mol, radius=2, n_bits=2048):
    """
    ?$?ECFP4 (Extended-Connectivity Fingerprints)
    ???:
        mol: RDKit??????
        radius: ?????? (2=ECFP4, 4=ECFP6)
        n_bits: ?????? (2048=???)
    ???:
        BitVec: Morgan????????n    """
    if mol is None:
        return None
    fp = AllChem.GetMorganFingerprintAsBitVect(mol, radius, nBits=n_bits)
    return fp

# ????$?
fingerprints = {}
for compound in compound_library:
    mol = Chem.MolFromSmiles(compound['smiles'])
    fingerprints[compound['id']] = calculate_ecfp4(mol)
```

**?????????**:
| ?????? | ??? | ??? | ?????? |
|---------|------|------|---------|
| ECFP2 | 1024/2048 | ?????????????????| ????????????|
| ECFP4 | 1024/2048 | ???2?????? | ??????????? |
| ECFP6 | 1024/2048 | ???3???????????| ????????|
| MACCS | 166 | ???????????? | ??'?????? |

#### 3.2 ????????n
```python
from rdkit import DataStructs

def calculate_tanimoto(fp1, fp2):
    """
    ?$?Tanimoto?????(Jaccard???)
    ???: |A ??B| / |A ??B|
    ???: [0, 1]
    1 = ??????, 0 = ??????
    """
    if fp1 is None or fp2 is None:
        return 0.0
    try:
        return DataStructs.TanimotoSimilarity(fp1, fp2)
    except:
        return 0.0
```

**???????*:
- ??????: T > 0.85
- ??????: 0.5 < T < 0.85
- ????? T < 0.5

#### 3.3 ???????????n
```python
def ligand_based_screening(compound_library, known_actives, fingerprint_type='ecfp4'):
    """
    ?????????????????
    ???:
        compound_library: ???????????? [{'id': str, 'smiles': str}]
        known_actives: ?????????????? [{'id': str, 'smiles': str}]
        fingerprint_type: 'ecfp4' ??'maccs'
    ???:
        ??????????????[{'id', 'smiles', 'max_similarity', 'best_match'}]
    """
    results = []
    # ?$?????????????????n    active_fps = {}
    for active in known_actives:
        mol = Chem.MolFromSmiles(active['smiles'])
        fp = calculate_ecfp4(mol) if fingerprint_type == 'ecfp4' else calculate_maccs(mol)
        active_fps[active['id']] = fp
    # ?$???????????????n    for compound in compound_library:
        mol = Chem.MolFromSmiles(compound['smiles'])
        lib_fp = calculate_ecfp4(mol) if fingerprint_type == 'ecfp4' else calculate_maccs(mol)
        max_sim = 0.0
        best_match = None
        for active_id, active_fp in active_fps.items():
            sim = calculate_tanimoto(lib_fp, active_fp)
            if sim > max_sim:
                max_sim = sim
                best_match = active_id
        results.append({
            'id': compound['id'],
            'smiles': compound['smiles'],
            'max_similarity': max_sim,
            'best_match': best_match,
            'fingerprint': lib_fp
        })
    results.sort(key=lambda x: x['max_similarity'], reverse=True)
    return results
```

---

### ???4: ?????????????(??????)

#### 4.1 AutoDock Vina ??????

**???AutoDock Vina**:
```bash
# ?????????
vina --receptor protein.pdbqt \
     --ligand compound.pdbqt \
     --center_x 10.5 --center_y 25.3 --center_z 42.1 \
     --size_x 20 --size_y 20 --size_z 20 \
     --exhaustiveness 32 \
     --num_modes 10 \
     --out results.pdbqt
```

**????????????** (vina_config.txt):
```
receptor = protein.pdbqt
ligand = ligands.pdbqt
center_x = 10.5
center_y = 25.3
center_z = 42.1
size_x = 20
size_y = 20
size_z = 20
exhaustiveness = 32
num_modes = 10
energy_range = 4
out = docking_results.pdbqt
log = docking_log.txt
```

#### 4.2 ???????h?

**Vina??????**:
```
?G = ?G_target + ?G_ligand + ?G_conf + ?G_torsion + ?G_clash + ...
     + ??? + ????????? + ?????? - ?????????????
```

**?????????**:
| ?????? | Vina??? (kcal/mol) | ??? |
|---------|-------------------|------|
| ?????| < -10 | ???????????|
| ?????? | -8 to -10 | ???????????|
| ?????| -6 to -8 | ???????|
| ?????| > -6 | ??????????|

---

### ???5: ?????? (Consensus Scoring)

#### 5.1 ????????n
```python
def normalize_scores(scores, method='minmax'):
    """???????????[0, 1]???"""
    if method == 'minmax':
        min_s = min(scores)
        max_s = max(scores)
        if max_s == min_s:
            return [0.5] * len(scores)
        return [(s - min_s) / (max_s - min_s) for s in scores]
```

#### 5.2 ?????????

**???A: ??????** (???)
```python
def consensus_score_weighted(similarity_scores, docking_scores, w_sim=0.4, w_dock=0.6):
    """
    ?????????
    ???:
        similarity_scores: ????????[0, 1]
        docking_scores: ?????? (???????????)
        w_sim: ????????n        w_dock: ?????????
    """
    # ????????? (?????? -12 ??0)
    dock_norm = [max(0, (s + 12) / 12) for s in docking_scores]
    consensus = []
    for sim, dock in zip(similarity_scores, dock_norm):
        score = w_sim * sim + w_dock * dock
        consensus.append(score)
    return consensus
```

**?????????**:
| ??? | w_similarity | w_docking | ??? |
|------|-------------|-----------|------|
| ???????| 0.6 | 0.4 | ?????????????? |
| ???????| 0.3 | 0.7 | ????????? |
| ??? | 0.4 | 0.6 | ?????? |

#### 5.3 ??????????n
```python
def generate_final_ranking(compounds, similarity_scores, docking_scores):
    """??????????????????????"""
    n = len(compounds)
    # 1. ????????n    sim_norm = normalize_scores(similarity_scores, 'minmax')
    dock_norm = normalize_scores(docking_scores, 'minmax')
    # 2. ?$???????
    consensus = consensus_score_weighted(sim_norm, docking_scores, w_sim=0.4, w_dock=0.6)
    # 3. ??????
    results = []
    for i, compound in enumerate(compounds):
        results.append({
            'rank': i + 1,
            'id': compound['id'],
            'smiles': compound['smiles'],
            'similarity': similarity_scores[i],
            'docking_score': docking_scores[i],
            'consensus_score': consensus[i]
        })
    # 4. ???????????n    results.sort(key=lambda x: x['consensus_score'], reverse=True)
    # 5. ??????
    for i, r in enumerate(results):
        r['final_rank'] = i + 1
    return results
```

---

## ??????: ???????g?

### execute.py ??????

```bash
# ??????
python execute.py --target protein.pdb --library compounds.smi --output results/

# ????????????
python execute.py \
    --target protein.pdb \
    --library compounds.smi \
    --known_actives actives.smi \
    --output results/

# ?????????
python execute.py \
    --target protein.pdb \
    --library compounds.smi \
    --binding_site binding_site.json \
    --output results/

# ??????
# --target, -t:       ????????? (PDB???)
# --library, -l:      ????????? (SMILES???)
# --known_actives, -k: ??????????? (SMILES???, ????
# --binding_site, -b: ????????? (JSON???, ????
# --output, -o:      ?????? (???: outputs/)
# --top_n, -n:       ?????op hits??? (???: 100)
```

### ??????

```bash
# ??????
pip install rdkit-pypi

# ???????(?????????)
pip install meeko pymol openbabel
```

---

## ??????: ?????????

### 3.1 ????????(screening_results.json)

```json
{
  "metadata": {
    "target": "protein.pdb",
    "library": "compounds.smi",
    "total_compounds": 1000,
    "filtered_compounds": 850,
    "timestamp": "2026-04-01T10:00:00Z"
  },
  "screening_summary": {
    "method": "consensus",
    "weights": {"similarity": 0.4, "docking": 0.6},
    "top_hits_count": 10
  },
  "top_hits": [
    {
      "rank": 1,
      "compound_id": "COMPOUND_001",
      "smiles": "CC(=O)Oc1ccc(cc1)CC(NCCc2ccc(Cl)cc2)=O",
      "descriptors": {
        "mw": 432.3,
        "logp": 3.2,
        "hbd": 1,
        "hba": 3,
        "tpsa": 55.4
      },
      "similarity_score": 0.85,
      "docking_score": -9.2,
      "consensus_score": 0.78
    }
  ]
}
```

### 3.2 ????????? (all_compounds.csv)

```csv
rank,compound_id,smiles,mw,logp,hbd,hba,tpsa,similarity,docking,consensus
1,COMPOUND_001,CC(=O)Oc1ccc...,432.3,3.2,1,3,55.4,0.85,-9.2,0.78
2,COMPOUND_002,c1ccc(C(=O)N...,287.3,2.5,2,2,52.1,0.72,-8.8,0.72
```

---

## ??????: ?????????

### ????????n```
[ ] ???????????? (????????
[ ] ??????pLDDT > 70 ???????< 3?
[ ] ??????????g'
[ ] ??????????????n[ ] SMILES????g'
```

### ????????n```
[ ] ????????? (???80%?????
[ ] ???????????? (-12 ??-4 kcal/mol)
[ ] ??????????????(0 ??1)
[ ] ?????????????n[ ] ???????????????????n```

---

## ??????: ???????????n
### Q1: RDKit??????
```bash
# Windows
pip install rdkit-pypi
# Linux
pip install rdkit
# macOS
conda install -c conda-forge rdkit
```

### Q2: ?????????????n- ???????????????????n- ???????????????????n- ????????????????n
### Q3: ????????????
```python
# ????MILES?????nfrom rdkit import Chem
smiles = "CC(=O)Oc1ccc(cc1)CC(NCCc2ccc(Cl)cc2)=O"
mol = Chem.MolFromSmiles(smiles)
if mol is None:
    print("Invalid SMILES")
```

### Q4: ??????
```bash
# ??????????????npython execute.py --library large_library.smi --batch_size 1000 --output batch_results/
```

---

## ??????: ???????n
### ??????
1. **DrugCLIP**: Gao et al., "Contrastive Protein-Molecule Representation Learning", arXiv:2310.06367
2. **DeepDTA**: ?zt?rk et al., "Deep drug-target binding affinity prediction", Bioinformatics, 2019
3. **AutoDock Vina**: Eberhardt et al., "AutoDock Vina 1.2.0", Aust J Chem, 2021

### ??????
4. **RDKit**: Landrum et al., "RDKit: Open-source cheminformatics", J Cheminform, 2013
5. **Fpocket**: Le Guilloux et al., "Fpocket", BMC Bioinformatics, 2009
6. **P2Rank**: Kriv?k & Hoksza, "P2Rank", Bioinformatics, 2018

### ????????n7. **Lipinski**: "Experimental and computational approaches", Adv Drug Deliv Rev, 1997
8. **Ghose**: "A knowledge-based approach in designing", J Phys Chem, 1999
9. **Veber**: "Molecular properties that influence oral druglikeness", J Med Chem, 2002

---

## ???A: ??????

### ????????????
```
skill11-virtual-screening/test_inputs/
????? target.pdb              # ???????????n????? compound_library.smi    # ???????????(5??????)
????? known_actives.smi       # ???????????????
????? binding_site.json       # ?????????
```

### ??????
????????????:
```
outputs/
????? screening_results.json   # ??????????????
```

---

## ???B: ??????

### ???1: ?????a???????(HTS)
```python
# ????????????????????n# ???:
# 1. ??????????????????
# 2. ???faiss??????????????
# 3. ???top 5%?????????
```

### ???2: ????????n```python
# ????????????????????????ntargets = ['EGFR', 'SRC', 'ABL']
selectivity_scores = []
for compound in compounds:
    scores = [docking_score(compound, target) for target in targets]
    selectivity = max(scores) - min(scores)  # ?????= ???????n    selectivity_scores.append(selectivity)
```

### ???3: ADMET??????
```python
from rdkit.Chem import Descriptors

def predict_admet(mol):
    mw = Descriptors.MolWt(mol)
    logp = Descriptors.MolLogP(mol)
    tpsa = Descriptors.TPSA(mol)
    return {
        'intestinal_absorption': 'High' if tpsa < 140 else 'Low',
        'BBB_permeant': True if logp > 0 and tpsa < 90 else False
    }
```

---

*????g??? 1.0 | ??????? 2026-04-01*