A lot of drug candidates don’t fail because they “don’t work” (i.e. they’re not active). They fail because the body doesn’t cooperate, exposure is wrong, the molecule doesn’t behave as you’d expect, or the safety profile isn’t apparent until too late. On further investigation, it’s always the same problem: ADMET.
In plain English, ADMET data enables you to predict whether a compound is actually going to be a viable medicine (i.e., not just a very good molecule). In the current state of the art in the pipelines, ADMET data is not something you just tick a box for at the end of discovery. It is actually a layer of decision-making that will determine which projects you optimise, which ones you go on to test and which ones you kill off early. That is why ADMET in drug discovery has become a fundamental strategy rather than a peripheral activity.
This post by DrugsVault is a breakdown of what ADMET is, what constitutes ADMET data, how teams generate it, how it’s used throughout the process, and what “good ADMET” will look like in 2026.
ADMET is an acronym for five “key” questions that every drug candidate has to answer.
Think about permeability, solubility and whether it can actually get into the body at a usable level.
Does it remain in the plasma, bind very heavily to proteins or accumulate in tissues such as the liver or the brain?
The enzymes involved, the rate of the process and the potential hazards of the metabolites produced.
Renal, biliary or other routes and how long it lasts (half life).
Cytotoxicity, liver signals, cardiac risk, genotoxicity, off-target effects and the like.
The key point is that ADMET isn’t “preclinical only”. It’s looked at by the teams from the very start and is continually re-evaluated as the chemistry of the team progresses.
ADMET data is structured information (measured or predicted) that describes how a drug behaves in the body and how safe it may be.
It usually comes in three forms:

You’ll also hear people talking about curated collections being used to train the models, to benchmark the predictions and to screen the candidates. Those are typically packaged up as ADMET datasets.
ADMET is essentially a de-risking layer. It reduces the chance that you’ve spent months optimising potency only to find the compound can’t get to the target tissue, can’t be dosed, or has an unacceptable interaction profile.
What ADME/Tox affects:
You pick the compound that you’re going to develop into a drug, it’s not just the one with the best IC50.
Dissolution and elimination will have a major bearing on your decision as to whether you use a prodrug, a salt, a modified release drug or a new route.
Tox findings point to what you should go on to test and what you should record.
Fewer late-stage surprises means fewer expensive resets.
Most importantly, ADMET in drug discovery becomes a decision-making tool rather than a tick box. It helps the team to decide what to build, what to fix and what to stop.
ADMET shows up differently depending on stage:
Fast screening + triage. You’re doing the filtering and triaging of chemistry.
Deeper profiling, species translation, toxicology studies and exposure planning.
PK/PD monitoring, safety signals and DDI risk in the presence of real people.
Safety monitoring and signal detection.
Here are the main types of data to monitor, what is being measured and why it’s important.
What it measures: solubility, permeability, dissolution and bioavailability.
Why it matters: if absorption is poor you may never achieve therapeutic exposure.
Common indicators include:
Measurement: distribution and sites of accumulation of the compound.
Relevance: distribution affects both efficacy (will it get to the site of action?) and safety (does it accumulate in the wrong place?).
Typical measurements:
What it assesses: transformation and its pathways
Why it’s important: leads to clearance, variability, toxic metabolites and DDIs.
Common measurements:
What it measures: how the compound is cleared and how long it persists.
Significance: Excretion dictates the dosing frequency, the potential for accumulation and the degree of interpatient variability.
Commonly measured values:
Measurement: The safety hazards and the potential for adverse effects.
Significance: Toxicity markers could lead to the entire development programme being scrapped or could require such a low dose that the drug would be rendered useless.
Common areas:
Both are useful, just in different ways.
Best practice in 2026 looks like this:
Not all datasets are equally useful. A good ADMET dataset is one that is consistent, well-documented and traceable.
Checklist for strong dataset quality:
If you can’t explain the assay context and the origin of the data, then you can’t possibly use it for modelling or for any decision-making.
This is where ADMET becomes real rather than just theoretical.
ADMET is powerful, just messy in practice. Common issues include:
Solving these typically requires more than “better models” it requires better data ops.
If you want ADMET to be driving the decisions, then treat it like a product not a spreadsheet.
Best practices:

Potency gets the attention, but it’s ADMET that ultimately determines whether a compound can actually survive in the real world. If you treat ADMET as a continuous process, then you make faster decisions, you reduce the number of late-stage failures and you end up with actually developable candidates.
If there’s one thing to take away: good ADMET practice is not about gathering more data. It’s about gathering the right data, in a consistent format, with clear provenance and using that data to make decisions as early as possible.
PK (pharmacokinetics) focuses mainly on exposure over time (concentration, AUC, Cmax, half-life). ADMET is broader: it includes PK-related elements plus metabolism pathways, distribution behavior, excretion routes, and toxicity risk.
As early as hit-to-lead. Early ADMET screens help avoid optimizing potency on compounds that are unlikely to be developable, and they guide chemistry toward better real-world profiles.
Consistency and context. Datasets are most useful when assays are clearly defined, units are standardized, metadata is complete, and versioning/provenance are tracked so models can be trained and validated reliably.
Accelerate compound screening, improve lead optimization, and reduce late-stage development risks with structured, high-quality ADMET datasets that support more informed drug discovery decisions.