Drug-Likeness Rules Cheatsheet

Lipinski to PROTACs. The classic rules, the modern ones, and when to break them. Free, no signup.

Drug-likeness is a set of empirical rules that flag compounds with high oral-absorption risk. The thresholds emerged from statistical analysis of marketed drugs, they are heuristics, not biology. Use them as triage signals, not pass/fail gates.

Rules at a glance

Click any row to jump to the deep-dive card.

RuleYearParameters
Lipinski's Rule of 51997
MW ≤500LogP ≤5HBD ≤5HBA ≤10
Veber's Rules2002
Rotatable bonds ≤10TPSA ≤140 Ų(or HBD + HBA ≤12)
Egan's Rules2000
LogP -1 to 5.88TPSA ≤131.6 Ų
Ghose's Rules1999
MW 160-480LogP -0.4 to 5.6MR 40-130Atoms 20-70
QED2012
Continuous score 0 to 1Higher = more drug-like8 weighted properties
Lead-Likeness2004
MW ≤350LogP ≤3RotB ≤7
CNS MPO2010
MW ≤360ClogP ≤3cLogD7.4 ≤2TPSA 40-90 ŲHBD ≤1pKa ≤8.0

Why it exists

Pfizer's analysis of ~2,250 drugs that had entered Phase II trials. The Rule of 5 emerged as a pattern: orally bioavailable drugs tended to satisfy these four limits. The thresholds are heuristics, not biology. Lipinski's classic convention: HBD = count of N-H and O-H bonds; HBA = total count of N + O atoms. Modern SMARTS-based implementations (e.g. RDKit NumHAcceptors) refine this by excluding pyrrole NH, amide N, nitro N, and most guanidine-type N — reported HBA can therefore differ substantially from the strict N+O count, particularly for amide- or guanidine-rich molecules.

What violation actually means

A violation is a flag, not a disqualifier. ~35-40% of post-2000 oral approvals violate at least one Ro5 criterion and still get to market — closer to 40% in the 2013-2019 cohort, driven by MW inflation in oncology and antivirals (Stegemann 2023, Shultz 2019). The genuine red flag is two or more violations, which remains rare among approved orals (~3%).

Famous violators

Atorvastatin

MW 558 and ALogP ~5.0-5.4 — two violations, the supposed red flag. Yet among the top-selling oral drugs in history. The dihydroxy-heptanoic-acid tail is essential for HMG-CoA reductase binding.

Itraconazole

MW 705 and cLogP > 5. Oral antifungal, demonstrates that highly lipophilic compounds can still work given the right formulation.

Why it exists

GSK analysis of rat oral bioavailability for 1,100 compounds. Veber and colleagues found that rotatable bond count and polar surface area predicted oral absorption better than Lipinski's parameters in this dataset.

What violation actually means

Conformational flexibility (high RotB) and polarity (high TPSA) both raise the entropic and energetic cost of crossing membranes. Compounds breaching both limits are the highest-risk category.

Famous violators

Cyclosporin A

TPSA 279 Ų, well over Veber's limit. Orally bioavailable through chameleonic conformational shielding (covered in the bRo5 section below).

Why it exists

Pharmacopeia, Inc. statistical analysis of human intestinal absorption (HIA) data. Egan, Merz, and Baldwin originally used AlogP98 (≈ LogP) and 3D PSA (Cerius2, single low-energy conformer); modern implementations substitute Ertl's 2D TPSA, which correlates r ≈ 0.99 with 3D PSA, so the 131.6 Ų cutoff is applied to TPSA in practice.

What violation actually means

Compounds outside the Egan ellipse have a higher probability of poor passive absorption. Useful as an early-stage triage filter.

Famous violators

Atazanavir

TPSA ~171 Ų, exceeds Egan's 131.6 limit. Oral HIV protease inhibitor; absorption rescued by once-daily ritonavir boosting and pH-dependent solubility.

Why it exists

Knowledge-based filter derived from analysis of CMC drug database compounds. Ghose, Viswanadhan, and Wendoloski derived bounds capturing the qualifying-drug property space.

What violation actually means

Tighter than Lipinski. Useful when designing combinatorial libraries where you want to enrich for drug-like starting points rather than rule out terminal candidates.

Famous violators

Metformin

MW 129, MR ~37, ALogP ~ -1.2 — violates three of Ghose's four lower bounds (MW, MR, ALogP). Total atom count = 20 sits exactly at the lower bound (Ghose 1999 counts hydrogens). The canonical small-molecule example: drug-likeness filters can reject compounds for being too small or too polar, not just too large.

Why it exists

Bickerton, Paolini, Besnard, Muresan, and Hopkins replaced binary cutoffs with a continuous 0-1 score. Eight property distributions (MW, ALogP, HBD, HBA, PSA, RotB, aromatic rings, structural alerts) across 771 approved oral drugs are fit to asymmetric double-sigmoidal desirability functions; the QED score is their weighted geometric mean. Three weighting schemes are published; QED_w,mo (mean of optimal weight combinations) is the typical default in Bickerton 2012 and standard implementations (RDKit, silicos-it/qed).

What violation actually means

QED is a continuous score, not pass/fail. The instruction is "aim higher," not "clear a fixed threshold." For context, the paper ran a chemical-beauty survey where medicinal chemists rated compounds on attractiveness; QED on the chemist-rated attractive set averaged noticeably higher than on the unattractive set, confirming the score tracks expert intuition without supplying a universal cutoff. Caveats: trained on small-molecule oral drugs to 2012, so PROTACs, macrocycles, and oligonucleotides score artificially low; the aromatic-ring penalty is contested in fragment work; the structural-alerts list is heuristic.

Famous violators

Dasatinib

QED 0.47, moderate. BCR-Abl / Src inhibitor; three aromatic rings (phenyl, pyrimidine, thiazole) plus an aliphatic piperazine, combined with high HBA load — drag the unweighted score below the approved-drug average despite blockbuster clinical use.

Atazanavir

QED 0.15, very low. The same properties that violate Egan (MW 705, TPSA 171, RotB 18) collapse the geometric mean. Marketed only via once-daily ritonavir boosting.

Why it exists

Originally proposed by Teague, Davis, Leeson, and Oprea (1999) and elaborated by Hann and Oprea (2004): fragments and hits used as starting points for medicinal chemistry should be smaller and less lipophilic than final drugs, since optimisation tends to add MW and LogP.

What violation actually means

Lead-like is for screening libraries and starting points, not for end candidates. Filtering an HTS deck against lead-likeness keeps the optimisation runway open.

Famous violators

Ibuprofen

MW 206, LogP 3.97, already an optimised drug rather than a lead, useful as a marketed reference for where final compounds tend to land.

Why it exists

Pfizer's multi-parameter optimisation framework for CNS drug candidates, derived from analysis of marketed CNS vs non-CNS drugs. Each parameter scores 0-1; the sum (max 6) ranks candidates.

What violation actually means

CNS exposure depends on passive permeability across the BBB and on minimising P-gp efflux. Higher MPO scores correlate with lower clinical attrition for CNS targets. Wager 2010 recommends MPO ≥ 4 (out of 6) as the desirable cut-off; ~74% of marketed CNS drugs clear that bar versus ~60% of CNS clinical candidates.

Famous violators

Aripiprazole

Calculated MPO ≈ 3.6 (RDKit-based); published values cluster 3.6-4.0 across implementations depending on cLogP/cLogD predictor — borderline against the ≥ 4 threshold. Clinically successful atypical antipsychotic. CNS rules are a probabilistic ranking, not pass/fail gates.

Beyond Rule of 5

Where Lipinski stops, three concepts pick up

Lipinski reads the 2D structure. Big molecules need a 3D view. What the donors do in solution, what polarity the molecule actually shows, and how it reshapes between water and lipid. These decide whether a compound outside Ro5 still goes oral.

eHBD
Exposed donors

The H-bond donors solvent actually sees. Counted from solution NMR conformations, not from the 2D structure. Donors locked into intramolecular H-bonds do not count.

ePSA
Measured polar surface

The polarity the molecule presents in solution, read by SFC chromatography. Often disagrees with the Ertl sum from 2D atoms, especially for macrocycles and PROTACs.

Chameleonic
Adapts to environment

Big molecules can fold polar groups inward in lipid and expose them in water. One structure, two faces. This is how cyclosporin crosses membranes despite huge calculated TPSA.

Recent work puts numbers on chameleonic behavior: David 2021 uses molecular dynamics on 24 FDA-approved drugs plus a PROTAC to predict experimental chromatographic apparent polarity, defining "chameleonic efficiency indices" that flag bRo5 candidates likely to fold favourably in lipid environments. Matsson 2016 frames the broader cell-permeability landscape outside Ro5.

References: Matsson P et al. Adv Drug Deliv Rev 2016 ; David L et al. ChemMedChem 2021 .
Oral PROTACs

For oral PROTACs, the rule is eHBD ≤2

AstraZeneca's NMR-based analysis of clinical oral PROTACs (Schade 2024) derives eHBD ≤ 2 as the operational rule. The Arvinas dataset of ~1,800 rat-PK-profiled compounds (Pike 2023) independently shows that exposed donors dominate the absorption signal. Each exposed donor costs energy to strip on entry into the bilayer.

Worked example: Vepdegestrant

Vepdegestrant sits exactly at the eHBD = 2 boundary: the phenol OH (warhead) and the glutarimide NH (E3 ligand handle), both solvent-exposed. The rule is permissive at the edge — the molecule passes despite being well outside Ro5 on every other axis.

PROTAC vs Ro5, typical ranges
Ro5Oral PROTAC
MW≤500700-1000
LogP≤52-7
HBA≤10≤16
eHBD(N/A)≤2

Ranges from analyses of orally-dosed PROTACs in rat PK datasets. eHBD is the high-signal parameter. Lipophilicity, conformational rigidity, transporters, and formulation still matter.

Recent updates (2024)

Schade 2024 established eHBD ≤2 as the dominant oral-bioavailability predictor in clinical oral PROTACs across mouse, rat, and dog, and introduced eHBD / eHBA as experimental NMR-derived descriptors that capture solution behaviour better than the 2D Ertl sum. The same study placed the oral-PROTAC ePSA upper bound at ~170 Ų (much narrower than calculated tPSA). Scott 2024 reinforces that CRBN-engaging warheads (low MW, single HBD in the binding pharmacophore) consistently outperform VHL-engaging ones for oral PK. Linker rigidity, ring-rich and conformationally constrained, helps both degradation potency and metabolic stability.

References: Edmondson SD et al. Bioorg Med Chem Lett 2019 ; Pike A et al. Drug Discov Today 2020 ; Schade M et al. J Med Chem 2024 ; Scott JS et al. RSC Med Chem 2024 .
Use with judgement

When to break the rules

The Rule of 5 was fitted to small-molecule oral drugs in 1997. Plenty of marketed drugs since then violate it and work anyway. The pattern is consistent: violations are tolerable when the chemistry is dictated by mechanism, the formulation can rescue exposure, or a different framework (bRo5, PROTAC) governs the chemotype.

  • Atorvastatin. MW 558. Dihydroxyheptanoate tail is mechanistically required for HMG-CoA reductase inhibition; cannot be trimmed.
  • Cyclosporin A. MW 1202, TPSA 279 Ų. Chameleonic conformer hides polar groups intramolecularly during membrane crossing.
  • Lopinavir. MW 628 (over Lipinski), LogP ~5.9, 14 rotatable bonds (over Veber). Boosted with ritonavir to inhibit CYP3A4 and rescue oral exposure.
  • Vepdegestrant. Oral PROTAC, MW > 700. Passes eHBD ≤2, the parameter that actually correlates with PROTAC PK.