ADMET Optimization
Iteratively optimize absorption, distribution, metabolism, excretion, and toxicity properties of drug candidates.
ADMET Optimization
ADMET properties -- absorption, distribution, metabolism, excretion, and toxicity -- determine whether a potent compound can become a viable drug. This workflow uses complementary prediction tools to profile candidates, identify pharmacokinetic liabilities, and guide iterative structural modifications toward a balanced property profile.
Overview
Poor ADMET properties account for a significant proportion of clinical trial failures. Early and iterative ADMET assessment during lead optimization reduces late-stage attrition. This workflow uses ADMET-AI for high-throughput screening across the series and safety_classify for a safety verdict. ADMETLab3 and ProTox3 are retired and no longer part of it, so there is no second independent ADMET model to cross-check against; confirm the properties that decide a program with experimental assays.
Tool Chain
- ADMET-AI -- High-throughput batch ADMET prediction
safety_classify-- Safety verdict from antitarget screening, structural alerts and DDI potential
Prerequisites
- SMILES strings for your compound series (typically 5-50 analogs)
- Baseline activity data from docking or biochemical assays
- Understanding of target product profile (desired PK parameters)
Step-by-Step Guide
Establish the property baseline
Run your initial compound series through ADMET-AI to obtain an ADMET profile. Review each compound across absorption (Caco-2 permeability, HIA, P-gp substrate status), distribution (BBB penetration, plasma protein binding, volume of distribution), metabolism (CYP450 inhibition and substrate panels), excretion (clearance, half-life), and toxicity (hERG, AMES, DILI) endpoints, and note the liabilities.
Identify critical liabilities
Compile the most frequent red flags across your series. Common liabilities include high CYP3A4 inhibition (drug-drug interaction risk), low Caco-2 permeability (poor oral absorption), hERG channel inhibition (cardiac risk), and high clearance (short half-life requiring frequent dosing). Rank liabilities by severity relative to your target product profile. A CNS drug with poor BBB penetration is a showstopper; the same property is irrelevant for an intravenous oncology agent.
Scale the screen across the analog set
For larger analog sets (50+ compounds), submit the whole set to ADMET-AI in one batch and focus on the endpoints that match your identified liabilities. ADMET-AI processes thousands of compounds in minutes, enabling rapid SAR assessment across property dimensions.
Toxicity deep-dive
Review the ADMET-AI toxicity endpoints for your top candidates (DILI, hERG, AMES, ClinTox, Tox21) and run safety_classify for a safety verdict. No tool on the platform predicts an acute-toxicity class (LD50) or organ-specific toxicity scores, so those need experimental data.
Design structural modifications
Using the combined ADMET data, design targeted modifications to address specific liabilities. Common strategies include adding fluorine atoms to block metabolic soft spots, introducing polar groups to improve solubility, reducing lipophilicity to lower hERG risk, and capping metabolically labile amines. Generate modified analogs manually or use de novo design tools to explore the local chemical space.
Iterate and converge
Run modified analogs back through the same tool chain (ADMET-AI and safety_classify) and compare property profiles against the previous iteration. Track improvement trends across cycles using exported CSV data. Typically 3-5 design-predict-analyze cycles are sufficient to achieve a balanced ADMET profile without compromising target activity.
Expected Outputs
- Per-compound ADMET profiles from ADMET-AI
- Batch property matrices from ADMET-AI
- Safety verdicts from
safety_classify - Iteration-over-iteration property comparison tables
- Exportable CSV and PDF reports for team review
Tips and Best Practices
- Treat a single model's prediction as a prioritization signal, not a verdict. Validate the endpoints that decide a program experimentally.
- Establish a target product profile before starting optimization. Define acceptable ranges for each ADMET property based on your therapeutic area and route of administration.
- Track the potency-ADMET balance explicitly. A compound with perfect ADMET but 100-fold reduced potency is not an improvement.
- Use matched molecular pair analysis to isolate the property impact of each structural change.
- Flag compounds with a CAUTION or DANGEROUS
safety_classifyverdict for careful review, even if other ADMET properties look favorable. - Export iteration data as CSV to build internal SAR tables that can inform future programs on the same target class.
Run it from a prompt
The same goal as a copy-and-adapt recipe you can paste into a chat:
- Screen ADMET — one ADMET-AI screen, cross-checked with DeepPurpose.
- Triage a compound series — descriptor, filter, and ADMET-ranking prompts for a series.
- Filter for drug-likeness — apply medicinal-chemistry ADMET filters to a compound set.
Was this page helpful?