ADMET Properties

What are ADMET properties?

ADMET properties are the five fundamental pharmacokinetic and safety parameters evaluated for every drug candidate during discovery and development: Absorption, Distribution, Metabolism, Excretion, and Toxicity. Together, these five dimensions determine how a compound behaves in a living system — from the moment it is administered to its eventual elimination from the body.

ADMET properties answer the most fundamental questions about a drug candidate’s fitness for clinical development: Will it be absorbed into the bloodstream after oral administration? Will it reach the target tissue at sufficient concentration? How quickly will it be broken down by liver enzymes? How efficiently will it be cleared from the body? And will it cause harm to the heart, liver, or other organs at therapeutic doses?

Understanding ADMET properties is not just a regulatory checkbox — it is central to the science of medicinal chemistry and lead optimization. A compound can show excellent potency and selectivity against its molecular target in biochemical or cellular assays, yet still fail as a drug if it has poor oral bioavailability, is too rapidly metabolized to maintain therapeutic levels, or causes unacceptable off-target toxicity. The integration of ADMET properties assessment into early drug discovery — through computational prediction, high-throughput in vitro assays, and Structure-Activity Relationship (SAR) analysis — is one of the most important advances in modern pharmaceutical science.

ADMET properties data is a core component of Excelra’s GOSTAR SAR database, which contains experimental ADMET measurements across millions of compounds — providing the curated, high-quality data foundation for in silico ADMET prediction and SAR-guided optimization.

ADME vs. ADMET: What’s the difference?

ADME (Absorption, Distribution, Metabolism, Excretion) is the classical pharmacokinetic framework — describing the four processes that govern how the body handles a drug compound over time. These four processes together determine the concentration-time profile of a drug in plasma and tissues.

ADMET extends this framework by adding Toxicity as a fifth critical dimension. While ADME characterizes pharmacokinetics — what the body does to the drug — toxicity characterizes the drug’s safety impact on the body. In modern drug discovery, ADMET is the preferred term because safety assessment is now fully integrated with pharmacokinetic optimization rather than treated as a separate, later-stage activity.

The acronym DMPK (Drug Metabolism and Pharmacokinetics) is also commonly used in industry — particularly in pharmaceutical R&D departments — and is essentially synonymous with the ADME component of ADMET. DMPK scientists are responsible for designing and interpreting in vitro and in vivo ADMET studies from hit identification through to IND-enabling studies.

Why ADMET properties are critical in drug discovery

Poor ADMET properties are the leading cause of drug failure — responsible for approximately 40–50% of all clinical trial attrition across the pharmaceutical industry. An analysis of AstraZeneca’s internal drug discovery programs from 2005 to 2010 found that undesirable ADMET properties — including poor oral bioavailability, excessive metabolic clearance, drug-drug interaction risk, and cardiac (hERG) toxicity — were among the most common reasons for compound termination.

Historically, ADMET assessment was conducted late in the drug discovery process — after significant medicinal chemistry and biology investment had already been made in a compound series. When ADMET failures were discovered at this stage, they were costly and time-consuming to address. The shift toward early ADMET integration — applying in silico predictions and high-throughput in vitro assays during hit identification and hit-to-lead stages — has been one of the most impactful changes in pharmaceutical productivity over the past two decades.

The “fail fast, fail cheap” principle in drug discovery depends entirely on early ADMET assessment. Compounds with fatal ADMET liabilities — such as irreversible CYP450 inhibition, highly reactive metabolites, or hERG inhibition at sub-micromolar concentrations — are better identified and eliminated at the hit stage than after a full lead optimization campaign. This connects directly to Excelra’s cheminformatics and data curation services, which enable high-quality ADMET data generation and interpretation for drug discovery programs.

Distribution (D) — Definition & key ADMET parameters

Distribution describes how a drug is transported from the systemic circulation to tissues and organs throughout the body after absorption. Distribution determines whether a drug reaches its target tissue at a sufficient concentration — and whether it accumulates in non-target tissues in ways that could cause toxicity.

Volume of distribution (Vd)

Volume of distribution (Vd) is a theoretical parameter that reflects the extent of drug distribution from plasma into tissues. A high Vd (greater than 1 L/kg) indicates extensive tissue distribution; a low Vd (less than 0.1 L/kg) suggests the drug remains primarily in the plasma compartment. Vd is important for estimating loading doses, predicting drug accumulation, and understanding tissue exposure relative to plasma exposure.

Plasma protein binding (PPB)

Most drugs bind reversibly to plasma proteins — primarily albumin and α1-acid glycoprotein — and only the unbound (free) fraction is pharmacologically active. Highly protein-bound compounds (PPB greater than 99%) have a very small free fraction available to interact with the target. While high PPB alone does not prevent drug efficacy, it affects dose requirements, drug-drug interactions, and the relationship between plasma concentration and effect. PPB is measured using equilibrium dialysis or ultrafiltration.

Blood-Brain barrier (BBB) penetration

For CNS-targeted drugs, the ability to cross the blood-brain barrier is essential. BBB penetration is governed by physicochemical properties (lipophilicity, molecular weight, hydrogen bonding, polar surface area) and active transport. For CNS drugs, a brain-to-plasma ratio (Kp,brain) greater than 0.3 is generally desired. For non-CNS drugs, BBB penetration is often a liability — increasing CNS side effect risk — making low BBB penetration a desirable ADMET property in that context.

Metabolism (M) — Definition & key ADMET parameters

Metabolism — also called biotransformation — is the enzymatic transformation of a drug into metabolites, primarily in the liver. Metabolism can inactivate drugs (reducing duration of action), activate prodrugs (converting inactive forms to active drugs), or generate reactive toxic metabolites (a safety liability). Understanding and optimizing metabolic stability is central to ADMET optimization in lead development.

CYP450 metabolism & inhibition

The cytochrome P450 (CYP450) enzyme family — particularly CYP3A4, CYP2D6, CYP2C9, CYP2C19, and CYP1A2 — is responsible for metabolizing approximately 70–80% of all marketed drugs. Two key ADMET metabolism assessments are: (1) metabolic stability — whether the compound is rapidly cleared by CYP450 enzymes (measured as microsomal half-life); and (2) CYP450 inhibition — whether the compound inhibits CYP enzymes that metabolize co-administered drugs, creating drug-drug interaction (DDI) risk. Reversible, mechanism-based, or time-dependent CYP inhibition are all important ADMET safety flags.

Microsomal stability

Microsomal half-life (t½,mic) measures how quickly a compound is metabolized by liver microsomes — an in vitro surrogate for hepatic metabolic clearance. Compounds with short microsomal half-lives (less than 30 minutes) are predicted to have high hepatic extraction and short in vivo half-lives, potentially requiring frequent dosing. ADMET optimization for metabolic stability typically involves identifying the sites of oxidative metabolism through metabolite identification (met-ID) studies and modifying the scaffold to block or slow metabolism at those sites.

Reactive metabolites & covalent binding

Some drug metabolites are chemically reactive — containing electrophilic Michael acceptors, epoxides, or quinones — and can form covalent adducts with proteins and DNA. Reactive metabolite formation is a major ADMET toxicity liability associated with idiosyncratic drug reactions (IDRs), drug-induced liver injury (DILI), and hapten-mediated immunotoxicity. Early detection of reactive metabolite formation using glutathione (GSH) trapping assays or potassium cyanide (KCN) trapping is now standard in ADMET profiling.

Excretion (E) — Definition & key ADMET parameters

Excretion describes the elimination of a drug and its metabolites from the body — primarily through renal (kidney) and biliary (liver/gut) routes. Excretion determines the elimination half-life (t½), total body clearance (CL), and the duration of drug exposure.

Renal clearance

The kidneys eliminate water-soluble drugs and metabolites through glomerular filtration, active tubular secretion, and passive tubular reabsorption. Renal clearance is particularly important for renally dosed drugs (e.g., antibiotics, antivirals) and for compounds requiring dose adjustment in patients with renal impairment. Renal transporters — OAT1, OAT3, OCT2 — play an important role in active tubular secretion and can be sites of drug-drug interactions.

Hepatic clearance & Half-Life

Hepatic clearance reflects the liver’s capacity to extract and metabolize a drug from the portal blood. The elimination half-life (t½) — the time required for plasma concentration to fall by 50% — integrates volume of distribution and total body clearance into a single parameter governing dosing frequency. Drugs with short half-lives (less than 2 hours) typically require multiple daily doses unless formulated as extended-release preparations.

Transporters in excretion

Biliary efflux transporters — including P-gp (MDR1), BCRP, and MRP2 — mediate drug excretion into bile. In the gut, OATP transporters facilitate enterohepatic recirculation. Transporter-mediated excretion can significantly affect drug exposure, particularly for compounds with limited metabolic clearance. Transporter interaction panels are now routinely included in regulatory ADMET submissions per FDA/EMA guidance on drug interaction studies.

Toxicity (T) — Definition & key ADMET parameters

Toxicity — the fifth and most safety-critical component of ADMET properties — encompasses all adverse biological effects of a drug beyond its intended pharmacological action. Toxicity can arise from on-target effects at excessive concentrations, off-target pharmacology, or metabolite-mediated mechanisms. Early toxicity prediction is now recognized as essential for reducing late-stage clinical trial failures and post-marketing drug withdrawals.

hERG inhibition (cardiac toxicity)

The hERG (human Ether-à-go-go Related Gene) potassium channel — encoded by KCNH2 — regulates cardiac repolarization. Inhibition of hERG by drug compounds can cause QT interval prolongation, potentially leading to the life-threatening cardiac arrhythmia Torsades de Pointes (TdP). hERG inhibition is one of the most common causes of post-marketing drug withdrawals and is now screened in every drug discovery program at the hit-to-lead stage using patch-clamp electrophysiology or fluorescence-based binding assays. IC50 values below 1–3 μM for hERG inhibition are generally considered a significant ADMET toxicity liability.

Drug-Induced liver injury (DILI)

Drug-induced liver injury (DILI) is the most common serious drug safety event leading to regulatory action — including black-box warnings and market withdrawals. DILI can be predictable (dose-dependent, intrinsic) or idiosyncratic (rare, immune-mediated). Key ADMET parameters predicting DILI risk include: high lipophilicity (cLogP greater than 3), high daily dose, reactive metabolite formation, mitochondrial toxicity, and inhibition of the bile salt export pump (BSEP). In vitro DILI models — including primary human hepatocytes, 3D spheroids, and iPSC-derived hepatocytes — are increasingly used to stratify DILI risk in early ADMET profiling.

Genotoxicity (ames test)

Genotoxicity — the ability of a compound to damage genetic material — is assessed using the Ames test (bacterial reverse mutation assay), the in vitro micronucleus test, and the mouse lymphoma assay. A positive genotoxicity finding is typically a terminating ADMET flag in most drug discovery programs, as genotoxic compounds carry unacceptable risk of carcinogenicity and mutagenicity in humans.

Phototoxicity, phospholipidosis & other ADMET toxicity flags

Additional toxicity endpoints assessed in comprehensive ADMET profiling include: phototoxicity (UV-induced skin damage, particularly for compounds absorbing UV light); phospholipidosis (amphiphilic compounds accumulating in lysosomes — flagged by the cationic amphiphilic drug / CAD structure alert); skin sensitization; and thyroid disruption. These ADMET endpoints are increasingly included in regulatory submissions as part of the ICH safety guidelines (S7A, S7B, E14) for non-clinical safety pharmacology.

Lipinski’s rule of five & physicochemical ADMET properties

Lipinski’s Rule of Five (Ro5), published by Christopher Lipinski of Pfizer in 1997, is the most widely cited framework for predicting oral drug-likeness from molecular physicochemical properties. The Ro5 states that an orally bioavailable drug-like molecule should satisfy most of the following criteria:

• Molecular weight (MW) ≤ 500 Da
• Calculated partition coefficient (cLogP) ≤ 5
• Number of hydrogen bond donors (HBD) ≤ 5
• Number of hydrogen bond acceptors (HBA) ≤ 10
• Topological polar surface area (TPSA) ≤ 140 Ų (added later as a fifth criterion)

These physicochemical parameters are directly related to the Absorption component of ADMET properties — compounds violating two or more Ro5 criteria are statistically less likely to be orally absorbed. The Ro5 is a powerful early filter for virtual screening of compound libraries but has important limitations: it applies only to compounds absorbed by passive transcellular diffusion and does not account for active transport mechanisms. Many successful oral drugs — including several macrolide antibiotics, HIV protease inhibitors, and emerging macrocycle drugs — violate one or more Ro5 criteria.

Beyond rule of five (bRo5)

Modern drug discovery increasingly targets challenging protein-protein interaction (PPI) surfaces and previously undruggable targets — requiring larger, more complex molecular scaffolds that fall outside Ro5 space. PROTACs, macrocycles, and stapled peptides — all exceeding 500 Da in molecular weight — represent the “beyond Rule of Five” (bRo5) frontier of drug discovery. ADMET assessment in bRo5 space requires specialized assays (e.g., equilibrated Caco-2 protocols optimized for high-MW compounds) and updated in silico models trained specifically on bRo5 compound data. This is an area where GOSTAR’s diverse ADMET dataset — including data on PROTACs and TPD compounds in GOSTAR TPD — provides particular value.

In Silico ADMET prediction methods

In silico ADMET prediction uses computational models to predict ADMET properties directly from chemical structure — enabling rapid, cost-efficient ADMET screening of virtual compound libraries before synthesis. This has become one of the most transformative advances in drug discovery efficiency over the past decade.

QSPR / QSAR models

Quantitative Structure-Property Relationship (QSPR) models correlate molecular descriptors — calculated from 2D or 3D chemical structure — with experimentally measured ADMET endpoints. Classical molecular descriptors include topological descriptors, electrostatic surface descriptors, and fingerprints. QSPR models built on high-quality, structurally diverse training data — such as GOSTAR’s curated ADMET datasets — deliver superior predictive accuracy compared to models trained on smaller or less diverse databases. Excelra’s published study comparing GOSTAR vs. benchmarking datasets for Caco-2 permeability prediction demonstrated this advantage directly: GOSTAR-trained models outperformed ChEMBL-trained models due to 2–7× greater structural diversity in key ADME parameters.

Widely used in silico ADMET tools

Several software tools and web servers are widely used for in silico ADMET prediction in drug discovery:

Physiologically based pharmacokinetic (PBPK) modeling

PBPK modeling integrates in vitro ADMET data — permeability, protein binding, metabolic clearance, transporter kinetics — into mechanistic compartmental models that simulate drug concentration-time profiles in specific tissues and organs. PBPK models are increasingly required by FDA and EMA for predicting drug-drug interactions, pediatric dosing, and the impact of organ impairment (renal, hepatic) on drug exposure. They connect directly to Excelra’s Quantitative Systems Pharmacology (QSP) and iQSP capabilities.

ADMET-Guided lead optimization

ADMET-guided lead optimization uses systematic SAR analysis to improve deficient ADMET properties while maintaining or improving potency and selectivity. This is the core activity of medicinal chemistry during the lead optimization stage of drug discovery — balancing ADMET properties against pharmacological activity in a multi-parameter optimization (MPO) framework.

Multi-Parameter optimization (MPO)

MPO scorecards integrate multiple ADMET and pharmacological parameters — potency, selectivity, solubility, permeability, metabolic stability, hERG — into a single composite score that guides compound prioritization and optimization. Pfizer’s CNS MPO score and AstraZeneca’s AZ-ADMET scorecard are examples of industry-standard MPO frameworks. Excelra’s GOSTAR platform supports MPO analysis through its integrated SAR and ADMET data, enabling medicinal chemists to visualize ADMET-activity trade-offs across compound series.

Common ADMET optimization strategies

Medicinal chemistry strategies for ADMET optimization include:

• Improving metabolic stability — blocking oxidative metabolism sites with fluorine, deuterium, or methyl groups; replacing metabolically labile functional groups (e.g., N-methyl, benzylic positions)
• Reducing hERG liability — introducing polar groups to reduce lipophilicity; reducing basic nitrogen content; increasing conformational rigidity to disfavor hERG binding
• Improving aqueous solubility — adding ionizable groups (acids, bases); introducing polarity; using salt forms or prodrug strategies
• Reducing P-gp efflux — reducing molecular weight and hydrogen bond count; introducing rigidity; using prodrug approaches for CNS delivery
• Eliminating reactive metabolite formation — replacing structural alerts (anilines, hydrazines, thiophenes, Michael acceptors) with safer bioisosteres

SAR analysis with Structure-Activity Relationship data from GOSTAR directly enables these optimization strategies by revealing how structural modifications in analogous compound series have impacted ADMET endpoints.

AI & machine learning in ADMET property prediction

Artificial intelligence and machine learning are rapidly transforming in silico ADMET prediction — enabling more accurate, comprehensive, and interpretable ADMET property models than classical QSPR approaches.

Graph neural networks (GNNs) for molecular ADMET prediction

GNNs operate directly on molecular graphs — where atoms are nodes and bonds are edges — learning hierarchical representations of molecular structure without requiring pre-calculated descriptors. GNN-based ADMET models (Attentive FP, MPNN, D-MPNN) have demonstrated state-of-the-art performance on benchmark ADMET datasets, particularly for complex endpoints such as hERG inhibition, CYP450 inhibition, and DILI. Attention-based GNNs provide atom-level interpretability — highlighting the specific structural features driving ADMET predictions.

Transformer & foundation models for ADMET

Large pre-trained molecular foundation models — including ChemBERTa, MolBERT, and Uni-Mol — learn rich molecular representations from millions of unlabeled molecules through self-supervised pre-training. Fine-tuning these foundation models on ADMET property datasets enables high-accuracy ADMET prediction even with limited labeled training data — particularly valuable for rare ADMET endpoints with sparse experimental measurements. The PharmaBench benchmark (2024) demonstrated that LLM-enhanced ADMET models significantly outperform classical QSPR approaches on multiple ADMET endpoints. Excelra’s AI/ML capabilities span these approaches.

Generative AI for ADMET-Optimized molecule design

Generative AI models — variational autoencoders (VAEs), generative adversarial networks (GANs), and diffusion models — can generate novel molecular structures with specified ADMET property profiles. By conditioning molecular generation on target ADMET endpoints (e.g., high Caco-2 permeability, low hERG inhibition, high microsomal stability), generative models accelerate ADMET optimization by proposing structurally diverse candidates that computationally satisfy the required ADMET profile before synthesis. This approach is increasingly used in conjunction with GOSTAR SAR data to guide generative design within validated chemical space.

GOSTAR & ADMET data in drug discovery

Excelra’s GOSTAR SAR database is one of the most comprehensive curated ADMET data resources available to the pharmaceutical industry — and a key differentiator from competing databases such as ChEMBL.

A direct comparison study by Excelra — GOSTAR vs. ChEMBL: Data diversity and compound coverage — demonstrated that GOSTAR contains 2–7× more unique chemical structures for key ADME parameters (including Caco-2, solubility, microsomal stability, plasma protein binding) than ChEMBL, when assessed by Tanimoto fingerprint similarity analysis. This structural diversity is crucial for training in silico ADMET models with high predictive accuracy across diverse chemical scaffolds — because model generalizability depends directly on the diversity of the training data.

GOSTAR’s ADMET data coverage includes experimental measurements for:

• Caco-2 permeability (Papp A→B, B→A, efflux ratio) — over 10,000 curated compounds
• Aqueous solubility (kinetic and thermodynamic)
• Plasma protein binding (PPB %)
• Microsomal stability (human, rat, mouse liver microsomes)
• CYP450 inhibition (CYP3A4, CYP2D6, CYP2C9, CYP2C19, CYP1A2)
• hERG inhibition (IC50, patch clamp)
• P-glycoprotein (P-gp) efflux ratio
• Blood-brain barrier (BBB) penetration (Kp,brain)
• In vitro and in vivo clearance values

For drug discovery teams building custom ADMET prediction models, optimizing lead series ADMET profiles, or conducting SAR-driven ADMET analysis, GOSTAR provides the curated, high-quality experimental data foundation that makes the difference between adequate and exceptional predictive model performance.

How Excelra supports ADMET optimization

Excelra provides integrated ADMET data, computational tools, and expert scientific support to drug discovery teams — from early virtual screening through IND-enabling ADMET characterization.

• GOSTAR ADMET database — access to 2–7× more structurally diverse ADMET data than ChEMBL across all key ADME parameters; curated experimental data for building and validating in silico ADMET models. Request GOSTAR access
• Cheminformatics & ADMET modeling — custom QSPR and deep learning ADMET prediction models built on GOSTAR data; multi-parameter optimization (MPO) scoring; structure-activity relationship analysis integrating ADMET endpoints. See Excelra’s cheminformatics services.
• SAR-Guided ADMET optimization — systematic SAR analysis across compound series to identify structural modifications improving metabolic stability, reducing hERG liability, and optimizing oral bioavailability. Powered by Structure-Activity Relationship (SAR) data from GOSTAR.
• Caco-2 Permeability prediction — published benchmarking study demonstrating GOSTAR-trained models outperform industry benchmark datasets for Caco-2 prediction; see our Caco-2 permeability blog.
• PBPK & QSP modeling — integration of in vitro ADMET data into PBPK and Quantitative Systems Pharmacology (QSP) models for human PK prediction, DDI assessment, and dose optimization.
• Drug Discovery data curation — expert data curation services for ADMET datasets, ensuring the data quality needed for reliable predictive ADMET modeling.

Conclusion

ADMET properties sit at the heart of modern drug discovery. Every drug candidate — no matter how potent or selective — must demonstrate acceptable absorption, distribution, metabolism, excretion, and toxicity profiles before it can deliver therapeutic benefit to patients. The approximately 40–50% of clinical trial failures attributable to poor ADMET properties represent a staggering cost to the pharmaceutical industry — and a reminder that potency alone is never enough.

The integration of ADMET assessment earlier in the drug discovery process — through high-quality in silico prediction, high-throughput in vitro screening, and SAR-guided medicinal chemistry optimization — is one of the most impactful strategies for improving pharmaceutical R&D productivity. AI and machine learning are now accelerating this integration further: GNN-based ADMET models, molecular foundation models, and generative AI tools for ADMET-optimized compound design are collectively pushing the boundary of what can be predicted computationally before synthesis.

The quality of ADMET data that underlies these computational models is the critical variable that separates good predictions from great ones. GOSTAR’s curated, structurally diverse ADMET dataset — containing 2–7× more unique compound structures for key ADME parameters than publicly available alternatives — provides the high-quality experimental foundation that powers superior in silico ADMET models for drug discovery teams worldwide.

For organizations seeking to improve the ADMET profiles of their compound libraries, reduce attrition in their drug discovery pipelines, or build best-in-class predictive ADMET models, Excelra’s combination of GOSTAR data, cheminformatics expertise, and SAR-driven optimization capabilities offers a comprehensive, evidence-based solution.