ADMET predictions and measured performance

Permeability, distribution, metabolism and safety screening in one workflow. Quantitative outputs and chemical context support compound prioritization.

Measured ADMET performance

Strong results across complementary endpoints and evaluation settings.

DILI risk discrimination

0.989

AUROC · 96 public test compounds

Identity, fingerprint and scaffold references excluded.

0.952 with identity/fingerprint exclusions only; same 96 cases.

Caco-2 permeability

0.281

MAE, log Papp · 182 compounds

Current alternative physical-mode evaluation on all 182 compounds in the known public development panel.

The public panel informed development. The default ADMET profile uses the hybrid route; its MAE is 0.914 under its separately reported reference-exclusion protocol.

CYP3A4 inhibition

0.839

AUROC · 2,467 test compounds

Reference identities, fingerprints and scaffolds excluded. AUPRC: 0.814.

CYP2C9 AUROC 0.835; CYP2D6 0.821. All endpoint metrics and cohorts below.

Intrinsic clearance

CLint

Clearance and stability estimates

Estimate intrinsic clearance and metabolic-stability class as part of the ADMET workflow.

Assay-specific microsome measurements and reference-exclusion policies are reported in the endpoint table.

Blood-brain barrier

86.03%

Classification accuracy · current panel evaluation

6,713 correct among 7,803 scored; 7,807 requested with four failures. Current system evaluated on the reference panel.

Panel self-exclusion protocol; some equivalent chemical identities can contribute. Reference-excluded fixed-test accuracy: 82.02%.

hERG inhibition

0.804

AUROC · 8,879 compounds

Current reference-panel evaluation also achieves MAE 0.483 pIC50, with all cases scored.

Literal self-exclusion can retain equivalent chemical identities. Reference-excluded fixed-test AUROC: 0.775 on 132 cases.

FluxMateria supports permeability, distribution, metabolism and safety screening with quantitative predictions and chemical context. Each result above names its evaluated mode, cohort and metric. Reference panels, development panels and reference-excluded tests answer different questions; their results are reported with those scopes.

Endpoint performance

EndpointMeasured resultConfiguration and evaluation
SolubilityMAE 1.363 logS; 1,997 casesComplete route; fixed public test. Test identities, fingerprints and scaffolds excluded. Two fallback results included.
Plasma protein bindingMAE 10.517 percentage points; all 559 casesPercent bound on a 0–100 scale. Fixed public test with identity/fingerprint/scaffold reference exclusions.
Intrinsic clearanceStandalone microsome Spearman 0.208 / 0.099; 221 casesIdentity exclusions / additional scaffold exclusions. The profile output under scaffold exclusions gives 0.059. Microsome results do not establish hepatocyte-assay accuracy.
Blood-brain barrier86.03% accuracy on 7,803 scored; four failures among 7,807 requestedCurrent reference-panel evaluation under self-exclusion; equivalent identities can contribute. Fixed reference-excluded test: 82.02% accuracy / 83.01% balanced accuracy on all 406.
hERG inhibitionAUROC 0.804; MAE 0.483 pIC50; all 8,879 panel casesCurrent reference-panel evaluation under literal self-exclusion; equivalent identities can contribute. Fixed reference-excluded test: AUROC 0.7751 standalone / 0.7753 profile, all 132 cases.
Caco-2 permeabilityAlternative mode MAE 0.281; production profile MAE 0.914Current evaluations on 182 cases. The alternative mode uses a known development panel. The profile evaluation excludes identities, fingerprints and test scaffolds from reference support.
DILIAUROC 0.952 / 0.989; the same 96 casesComplete v4.32 system, identity/fingerprint exclusions / additional scaffold exclusions. All cases scored on the known public test.
CYP inhibition2C9 / 2D6 / 3A4: AUROC 0.835 / 0.821 / 0.839; AUPRC 0.692 / 0.562 / 0.814Current fixed tests of 2,419 / 2,626 / 2,467 cases. Test identities, fingerprints and scaffolds excluded. Each isoform has its own endpoint and metric.

Capabilities available in the ADMET workflow

CapabilityOutput available
Absorption and permeabilityCaco-2 apparent permeability, permeability class and associated screening context.
DistributionBlood-brain barrier classification, aqueous solubility and plasma protein binding.
MetabolismIntrinsic-clearance estimates, metabolic-stability classification and per-isoform CYP inhibition predictions.
Safety screeninghERG inhibition estimates and DILI risk with mechanism and exposure context.

Evaluation settings

These are measurements of the currently evaluated system. Each product mode has its own input, cohort and reference policy. Reference-assisted panel evaluations and reference-excluded tests remain distinct.

AUROC, AUPRC, classification accuracy, MAE and rank correlation measure different aspects of prediction. TDC ranking comparisons require the specified endpoint, test and metric. Newly selected cases can test transfer to your application.

Evaluate FluxMateria for your application

Choose the properties, input conditions and scoring criteria for your application. A frozen system and partner-selected cases provide a direct product evaluation.

Request access Explore benchmarks