Drug-Likeness Scoring

Drug-likeness scores evaluate whether a molecule has properties consistent with known drugs. ChemAudit implements multiple established drug-likeness filters.

Available Filters

Filter	Criteria	Pass Rate in DrugBank
Lipinski (Ro5)	MW ≤ 500, LogP ≤ 5, HBD ≤ 5, HBA ≤ 10	~85%
QED	Quantitative Estimate (0-1 scale)	Mean ~0.6
Veber	Rotatable bonds ≤ 10, TPSA ≤ 140	~90%
Rule of Three	MW ≤ 300, LogP ≤ 3, HBD ≤ 3, HBA ≤ 3	Fragment-like
Ghose	MW 160-480, LogP -0.4-5.6, atoms 20-70, MR 40-130	~75%
Egan	LogP ≤ 5.88, TPSA ≤ 131.6	~85%
Muegge	Multiple criteria	~80%

Lipinski's Rule of Five

The most widely used drug-likeness filter:

Property	Criterion	Rationale
Molecular Weight	≤ 500 Da	Oral bioavailability correlation
LogP	≤ 5	Membrane permeability
H-Bond Donors	≤ 5	Solubility and permeability
H-Bond Acceptors	≤ 10	Solubility and permeability

Violations allowed: Up to 1 violation is often acceptable

Exceptions

Many successful drugs violate Lipinski: antibiotics, antifungals, natural products, and biologics-inspired molecules.

QED (Quantitative Estimate of Drug-likeness)

Continuous score from 0 to 1 based on 8 molecular properties:

• Molecular weight
• LogP
• H-bond donors
• H-bond acceptors
• Polar surface area
• Rotatable bonds
• Aromatic rings
• Structural alerts

QED Score	Interpretation
0.8-1.0	Highly drug-like
0.6-0.8	Moderately drug-like
0.4-0.6	Low drug-likeness
0.0-0.4	Not drug-like

Veber Rules

Focus on oral bioavailability:

Property	Criterion	Purpose
Rotatable Bonds	≤ 10	Molecular flexibility
TPSA	≤ 140 Ų	Membrane permeability

Developed from analysis of rat oral bioavailability data.

Rule of Three (Ro3)

Fragment-like criteria for screening libraries:

• MW ≤ 300 Da
• LogP ≤ 3
• H-bond donors ≤ 3
• H-bond acceptors ≤ 3

Used for fragment-based drug discovery.

Ghose Filter

Defines drug-like chemical space:

• MW: 160-480 Da
• LogP: -0.4 to 5.6
• Atom count: 20-70
• Molar refractivity: 40-130

Based on analysis of the World Drug Index.

Egan Rules

Simple bioavailability filter:

• LogP ≤ 5.88
• TPSA ≤ 131.6 Ų

Derived from 1,2000 compounds with known oral bioavailability.

Muegge Filter

Comprehensive 9-parameter filter for druglike chemical space:

Property	Threshold
Molecular Weight	200–600 Da
LogP	−2 to 5
TPSA	≤ 150 A²
Ring Count	≤ 7
Carbon Count	> 4
Heteroatom Count	> 1
Rotatable Bonds	≤ 15
H-Bond Donors	≤ 5
H-Bond Acceptors	≤ 10

Pass criterion: All 9 parameters must pass (0 violations).

Reference: Muegge et al. (2001). Simple selection criteria for drug-like chemical matter. Journal of Medicinal Chemistry, 44(12), 1841–1846.

Consensus Drug-Likeness (0–5)

A composite score counting how many of the major filter sets the molecule passes:

1. Lipinski (≤ 1 violation)
2. Veber (both parameters pass)
3. Egan (both parameters pass)
4. Ghose (all parameters in range)
5. Muegge (all 9 parameters pass)

Score	Interpretation
5	Excellent drug-likeness
4	Very good
3	Acceptable
2	Borderline
0–1	Poor drug-likeness

Lead-Likeness

Identifies molecules in lead-like chemical space, which provides room for optimization during hit-to-lead campaigns.

Property	Threshold
Molecular Weight	200–350 Da
LogP	−1 to 3
Rotatable Bonds	≤ 7

Pass criterion: All must pass.

API Usage

curl -X POST http://localhost:8001/api/v1/score \
  -H "Content-Type: application/json" \
  -d '{
    "molecule": "CC(=O)Oc1ccccc1C(=O)O",
    "include": ["druglikeness"]
  }'

Response:

{
  "druglikeness": {
    "lipinski": {
      "passed": true,
      "violations": 0,
      "mw": 180.16,
      "logp": 1.19,
      "hbd": 1,
      "hba": 4
    },
    "qed": {
      "score": 0.71,
      "interpretation": "Moderately drug-like"
    },
    "veber": {
      "passed": true,
      "rotatable_bonds": 3,
      "tpsa": 63.6
    },
    "ro3": {
      "passed": true,
      "violations": 0
    },
    "ghose": {
      "passed": true,
      "violations": 0
    },
    "egan": {
      "passed": true
    },
    "muegge": {
      "passed": true,
      "violations": 0
    },
    "interpretation": "Passes Lipinski and Veber rules"
  }
}

Interpreting Results

All Filters Pass

Excellent drug-likeness. Molecule has properties consistent with most known oral drugs.

Lipinski Passes, Others Fail

Reasonable drug-likeness. Review specific failures to understand limitations.

Multiple Failures

May still be drug-like if:

• Targeting specific therapeutic areas (antibiotics often fail)
• Natural product (often fail due to complexity)
• Prodrug or special formulation

Low QED (less than 0.4)

Indicates poor overall drug-likeness, but context matters. Some successful drugs have low QED.

Use Cases

Virtual Screening

Pre-filter compound libraries:

lipinski_passed = true AND
veber_passed = true AND
qed >= 0.5

Lead Optimization

Track drug-likeness during optimization:

• Monitor QED score changes
• Avoid introducing Lipinski violations
• Maintain favorable TPSA and rotatable bonds

Library Design

Design screening libraries with good drug-like properties:

• Ghose filter for general drug-like space
• Ro3 for fragment libraries
• Lipinski + Veber for lead-like compounds

Limitations

These rules are guidelines, not absolutes:

• Many drugs violate these rules
• Rules derived from oral drugs (not applicable to all routes)
• Don't account for specific mechanisms or targets
• Don't predict efficacy or safety

Context Matters

Use drug-likeness filters for prioritization, not as strict cutoffs. Always consider therapeutic area, target, and route of administration.

Best Practices

1. Use multiple filters: No single filter is perfect
2. Consider therapeutic area: Some areas (antibiotics, antivirals) often violate rules
3. Track QED over time: Monitor during optimization
4. Don't over-optimize: Chasing perfect drug-likeness can sacrifice potency
5. Validate experimentally: Predicted properties should be confirmed

References

1. Bickerton, G. R. et al. (2012). Quantifying the chemical beauty of drugs. Nature Chemistry, 4(2), 90–98.
2. Congreve, M. et al. (2003). A 'rule of three' for fragment-based lead discovery? Drug Discovery Today, 8(19), 876–877.
3. Egan, W. J. et al. (2000). Prediction of drug absorption using multivariate statistics. Journal of Medicinal Chemistry, 43(21), 3867–3877.
4. Ghose, A. K. et al. (1999). A knowledge-based approach in designing combinatorial or medicinal chemistry libraries. Journal of Combinatorial Chemistry, 1(1), 55–68.
5. Lipinski, C. A. et al. (2001). Experimental and computational approaches to estimate solubility and permeability. Advanced Drug Delivery Reviews, 46(1-3), 3–26.
6. Muegge, I. et al. (2001). Simple selection criteria for drug-like chemical matter. Journal of Medicinal Chemistry, 44(12), 1841–1846.
7. Veber, D. F. et al. (2002). Molecular properties that influence the oral bioavailability of drug candidates. Journal of Medicinal Chemistry, 45(12), 2615–2623.

Next Steps

• Scoring Overview — All scoring systems
• ADMET — ADMET predictions
• Safety Filters — Structural alerts