ever-works/orgs/admet-assessment

ADMET Assessment

Surface developability liabilities at hit stage, as first-class data, so a program does not commit to a series that will fail downstream.

Procedure

1. Frame the developability question. Route of administration and target profile set which ADMET properties actually matter for this program.
2. Estimate absorption and permeability from physicochemical properties and models; flag compounds unlikely to reach their target.
3. Assess metabolic liability. Predict metabolic soft spots and clearance risk; note reactive-metabolite potential.
4. Screen for toxicity flags. Structural alerts (toxicophores), cardiac (hERG) liability, and other class-known risks. Treat alerts as hypotheses to weigh, not automatic vetoes.
5. Consider distribution and exposure factors relevant to the target compartment (e.g., barrier penetration when needed, or exclusion when not).
6. Weigh predictions honestly. Computational ADMET is directional; state model confidence and the applicability domain rather than reporting point values as truth.
7. Rank liabilities by severity and stage. A liability that kills a program at the clinic outweighs a minor property miss.
8. Recommend which liabilities are series-defining, which are optimizable, and what experiment would resolve the biggest unknown.

Quality bar

• The profile is scoped to the program's route and target.
• Predictions carry confidence and applicability-domain caveats.
• Structural alerts are weighed, not treated as automatic pass/fail.
• Liabilities are ranked by severity and development stage.

Output format

An ADMET liability profile per compound or series: absorption, metabolism, toxicity flags, and distribution notes; each with predicted value, confidence, and a severity/stage rating, plus a recommendation and the highest-value confirming experiment.