Technical Bulletin · ADMET & PK Strategy

How to Build an ADMET Screening Strategy for Early Drug Discovery

A stage-gated framework for choosing which endpoints to predict, which to measure, and when evidence should escalate from rapid triage to decision-grade experimental support.

Discuss Your ADMET Strategy

Opening answer

Start with the decision, then scale the evidence

An effective early-discovery ADMET strategy applies endpoints according to the decision at the next gate, the number and structural diversity of compounds, the expected human exposure and route, and the consequence of being wrong. Hit discovery uses fast, inexpensive triage to identify solubility, permeability, stability, reactivity, or toxicity liabilities. Hit-to-lead adds reproducible experimental measurements for the liabilities most likely to limit exposure or interpretation. Lead optimization connects potency with clearance, distribution, metabolism, transporter, safety and formulation constraints in a multi-parameter design loop. Candidate selection requires an integrated, auditable evidence package, including repeat or orthogonal assays, in vivo pharmacokinetics and a justified path toward development-stage studies. Screening depth should therefore increase as compound count falls and decision consequence rises. Predictions rank and flag; experiments measure under defined conditions; in vivo studies establish system-level behavior; clinical implications integrate the exposure and therapeutic context.

Decision context

Three variables determine how deep the screen should go

Compound number and diversity

Hundreds or thousands of virtual or synthesized structures favor high-throughput calculations and a compact assay panel. Closely related analogues may support local structure–property reasoning, while novel chemotypes and unusual modalities are more likely to sit outside a model's applicability domain. A diverse series may need sentinel compounds selected across chemical space rather than testing only the most potent members.

Development stage and decision consequence

Early ranking tolerates more uncertainty because errors can be corrected cheaply. A nomination decision has asymmetric consequences: overlooking a major liability can delay development, while an overly conservative cutoff can discard a differentiable compound. As the cost of a wrong choice increases, confirmation should move from a single prediction or screen to orthogonal experimental systems, exposure-aware interpretation and in vivo evidence.

The intended route and dose matter throughout. A permeability signal can mean something different for a low-dose oral drug, a high-dose oral drug, or a non-oral product. Likewise, a CYP inhibition concentration has little clinical meaning without anticipated unbound exposure at the relevant site. Disease population, co-medications, organ impairment, duration of dosing and therapeutic margin may change which risk is decision-critical.

Stage progression

Move from rapid triage to integrated candidate evidence

1 · Hit Discovery

Remove obvious mismatches

Use structure quality checks, rule-based alerts, endpoint-specific models and limited high-throughput assays. Prioritize solubility, permeability or uptake relevant to the route, metabolic stability, basic CYP liability and broad cytotoxicity or reactivity alerts.

2 · Hit-to-Lead

Confirm tractable liabilities

Measure kinetic solubility, permeability or efflux, microsomal or hepatocyte stability, protein binding where it informs interpretation, CYP inhibition and early safety signals. Compare results within the same protocol and concentration context.

3 · Lead Optimization

Balance exposure and safety

Add intrinsic clearance, metabolite identification, reaction phenotyping as appropriate, transporter studies, time-dependent CYP inhibition or induction follow-up, cardiac ion-channel and broader safety assays, formulation-relevant solubility and exploratory in vivo PK.

4 · Candidate Selection

Support a defensible nomination

Integrate physicochemical, in vitro ADME, safety pharmacology signals, metabolite risk, dose projection and repeat-dose or species-relevant in vivo evidence. Define remaining uncertainties and development-stage studies.

Practical decision component

Stage-Gated ADMET Endpoint Matrix

"Predict/triage" means computational or rule-based prioritization; "measure" means an experimental result under a documented protocol; "deepen" means mechanistic, orthogonal or exposure-aware follow-up. Endpoint selection follows the modality, route, target product profile and known chemical-series risks.

Predict / rapid triageMeasure / confirmDeepen / integrate

Endpoint family	Hit Discovery	Hit-to-Lead	Lead Optimization	Candidate Selection	Key interpretation condition
Identity, structure & physicochemical properties	Structure standardization; ionization, lipophilicity and solubility estimates	pKa/logD and kinetic solubility for representative hits	Thermodynamic/formulation-relevant solubility; solid-state considerations as needed	Confirmed material form, purity and formulation-compatible property set	Salt, tautomer, stereochemistry, pH and solid form can change the apparent result.
Absorption & permeability	Route-relevant permeability/absorption prediction; structural alerts	Cell-based permeability and efflux screen	Mechanistic transporter follow-up or dissolution/absorption studies when indicated	Dose- and formulation-aware absorption assessment	Interpret with dose, solubility, ionization, transporter expression and assay recovery.
Distribution & binding	Plasma protein binding and volume-distribution estimates for ranking	Species-specific plasma protein binding if needed for clearance/exposure interpretation	Blood-to-plasma ratio, tissue distribution or brain penetration when relevant	Cross-species, unbound-exposure interpretation	Binding measurements must be reliable at relevant concentrations; total exposure is not unbound exposure.
Metabolic stability & clearance	Metabolic-stability prediction or compact microsomal screen	Microsomal/hepatocyte stability and intrinsic-clearance estimation	Species comparison, metabolite identification, enzyme contribution and IVIVE	Integrated clearance pathways and in vivo PK reconciliation	Assay system, nonspecific binding, extrahepatic clearance and transporter coupling affect translation.
CYP / enzyme DDI potential	Inhibition or metabolism-site flags for prioritization	Reversible inhibition screen for relevant major CYPs	IC50/Ki, time-dependent inhibition, induction and reaction phenotyping as triggered	Exposure-based static or PBPK assessment and development plan	Clinical DDI risk depends on unbound exposure, enzyme contribution, metabolites and dosing context.
Transporters	Substrate/inhibitor alerts where route or chemistry suggests relevance	Efflux or uptake assays for identified liabilities	Mechanistic substrate/inhibition studies for relevant transporters	Integrated victim/object and perpetrator/precipitant assessment	Cell system, expression, passive permeability and concentration range govern interpretation.
Safety and toxicity risk	Structural alerts, cytotoxicity and endpoint-specific predictions	Concentration-response cytotoxicity; selected genotoxicity or ion-channel screens	Orthogonal cardiac, genetic, hepatic or other target-organ assays based on risk	Integrated safety pharmacology and toxicology planning with exposure margins	Interpret alerts with biological context, concentration response and confirmatory evidence.
In vivo PK and exposure	Usually not routine	Optional cassette or focused studies when they answer a specific question	Exploratory single-dose PK, bioavailability and exposure–response linkage	Repeatable, species-appropriate PK and dose projection supporting nomination	Translate animal PK through species, formulation and human-exposure considerations.

Evidence hierarchy

Keep prediction, measurement and implication separate

Prediction

A model output, alert or estimated value conditional on the training data, representation, endpoint definition and applicability domain. It is used for ranking, hypothesis generation and resource allocation.

Experimental measurement

An observed assay result tied to protocol, biological system, concentration range, controls, analytical method and variability.

In vivo evidence

An integrated observation in an animal or other whole-system context that captures competing processes across absorption, distribution, metabolism and elimination.

Clinical or regulatory-support use

A conclusion linked to human exposure, intended population and a defined development question, supported by validation, documentation and the totality of evidence.

When evidence conflicts, first check identity and input quality, assay conditions, concentrations, recovery, cytotoxicity interference and whether the model endpoint matches the experiment. Then assess whether the molecule is inside the model's chemical and endpoint applicability domain and whether the biological endpoint aligns with the decision.

Prediction quality

Applicability domain, uncertainty and input quality must travel with every result

Verify the input

Document stereochemistry, protonation or ionization state, salt handling, mixture status and structure normalization. Incorrect molecular form can invalidate an otherwise suitable model output.

Locate the domain

Small-molecule models may not cover covalent compounds, metal-containing structures, macrocycles, peptides or degraders. Check chemical-space coverage and endpoint compatibility explicitly.

Quantify uncertainty

Combine local data density, method agreement, endpoint-specific validation and prediction intervals. A probability alone is not a complete confidence assessment.

Escalation rule: out-of-domain results, model conflicts, values near a decision threshold, and high-consequence liabilities should trigger experimental confirmation. External validation is more informative than random splits that leak close analogues. Ask: "Could this uncertainty change the current decision?"

Multi-parameter optimization

Avoid single-endpoint cutoffs and hidden trade-offs

Potency, selectivity, solubility, permeability, clearance, distribution and safety endpoints often pull chemical design in different directions. Preserve values, units, assay conditions and uncertainty rather than collapsing the profile into an opaque score.

Transparent trade-offs

Reserve hard stops for justified, decision-critical constraints. Desirability functions or Pareto views can reveal alternatives without implying that one weighting is biologically true.

Series versus compound

A scaffold-wide liability may justify redesign; one noisy result should prompt repeat testing. Keep chemical diversity until major liabilities are understood.

Auditable decisions

Record the question, compounds, evidence level, uncertainty, ranking logic, excluded alternatives and next experiments at each gate.

Experimental follow-up

Design confirmation around the failure mode

When a favorable prediction needs confirmation

The endpoint controls a high-consequence nomination or dosing assumption.
The compound is outside the applicability domain or structurally novel.
Predictions depend on uncertain ionization, solubility or molecular form.
The expected clinical or preclinical exposure approaches a biological effect concentration.
Regulatory-support use is anticipated and method documentation or validation is required.

When an unfavorable signal needs investigation

Check assay interference, aggregation, nonspecific binding, cytotoxicity and recovery.
Repeat with concentration response and appropriate positive/negative controls.
Use an orthogonal platform or biological system where feasible.
Test whether a metabolite, transporter or time-dependent mechanism explains the signal.
Compare against unbound exposure and therapeutic margin.

ICH M12 describes DDI evaluation as stepwise and tailored to the drug, intended population and therapeutic context. Its recommendations become especially relevant as a program approaches clinical development. FDA and EMA guidances likewise frame studies around defined development questions and exposure-informed evidence.

Implementation

A practical gate review

Define the next decision. State whether the team is removing weak hits, selecting analogues, choosing a lead or nominating a candidate.
Map consequence and reversibility. Identify which false negative or false positive would be most costly and whether chemistry can repair it.
Choose route- and exposure-relevant endpoints. Connect assays to the target product profile, plausible dose and intended population.
Assign an evidence level. Mark each result as predicted, experimentally measured, in vivo observed or clinically interpreted.
Set uncertainty rules. Define applicability-domain checks, orthogonal confirmation and escalation triggers.
Review the whole profile. Use transparent multi-parameter trade-offs, preserve diversity and document why compounds advance.
Plan the next evidence package. Pair the selection with a targeted plan to close the most decision-relevant gaps.

Relevant services

Connect the strategy to fit-for-purpose analysis

When internal screening reveals a prioritization or translation question, these existing services can support a defined portion of the workflow. Scope, inputs, endpoints and validation should be agreed for the specific project.

Request a Customized ADMET Plan

References

Scientific and Regulatory Sources

International Council for Harmonisation. ICH M12: Drug Interaction Studies. Final version, adopted 21 May 2024.
European Medicines Agency. Guideline on the Investigation of Drug Interactions, Revision 1.
U.S. Food and Drug Administration. Clinical Drug Interaction Studies — Cytochrome P450 Enzyme- and Transporter-Mediated Drug Interactions: Guidance for Industry. 2020.
Waring, M. J.; Arrowsmith, J.; Leach, A. R.; et al. An analysis of the attrition of drug candidates from four major pharmaceutical companies. Nature Reviews Drug Discovery 2015, 14, 475–486. doi:10.1038/nrd4609.
Kennedy, T. Managing the drug discovery/development interface. Drug Discovery Today 1997, 2, 436–444. doi:10.1016/S1359-6446(97)01099-4.
OECD. Principles for the Validation, for Regulatory Purposes, of (Q)SAR Models. Guidance on defined endpoints, algorithms, applicability domains, goodness-of-fit, robustness, predictivity and mechanistic interpretation.