ScreenServer — ADMET Demo Screen (30 approved/withdrawn drugs + known PAINS)

ADMET / molecular-property virtual screening · generated 2026-06-10 22:20 UTC · RDKit descriptors + 2048-bit Morgan fingerprints + gradient-boosted ensembles trained on MoleculeNet & Therapeutics Data Commons.

Triage summary (Claude)

Salbutamol (score 87) is the cleanest profile—negligible hERG (4%), CYP (≤15%), DILI (1%), and zero alerts—though its low caco2 (-5.45) flags permeability concerns consistent with poor oral absorption. Ibuprofen (81) and Morphine (79) are well-balanced with low CYP liabilities and good QED (0.82, 0.77), but Ibuprofen carries borderline DILI (45%) and Morphine notable CYP2D6 (34%) plus high BBB penetration. The single biggest liability across the set is concentrated hERG cardiotoxicity in the bottom tier: Ketoconazole (99%) and Loratadine (98%), both compounded by strong multi-CYP inhibition—Ketoconazole especially (CYP3A4 98%, CYP2C9 87%, DILI 99%, Ro5 violation).

Model held-out performance (both split protocols)

EndpointTaskSource Random splitScaffold splitNotes
Aqueous Solubility (logS)regMoleculeNetR²=0.900
RMSE 0.628
R²=0.854
RMSE 0.884
ESOL/Delaney. Higher = more soluble.
Lipophilicity (logD 7.4)regMoleculeNetR²=0.683
RMSE 0.668
R²=0.712
RMSE 0.675
AstraZeneca logD at pH 7.4. Sweet spot ~1-3.
BBB PenetrationclaMoleculeNetAUC=0.946
F1 0.94
AUC=0.799
F1 0.94
1 = crosses blood-brain barrier.
hERG Cardiotox (blocker)claTDCAUC=0.837
F1 0.90
AUC=0.904
F1 0.88
1 = hERG blocker (cardiotoxicity risk). Lower is safer.
Caco-2 PermeabilityregTDCR²=0.753
RMSE 0.405
R²=0.703
RMSE 0.437
Caco-2 apparent permeability. Higher = more permeable (> -5 good).
CYP3A4 InhibitionclaTDCAUC=0.917
F1 0.79
AUC=0.950
F1 0.82
1 = CYP3A4 inhibitor (drug-drug interaction risk).
CYP2D6 InhibitionclaTDCAUC=0.879
F1 0.61
AUC=0.901
F1 0.65
1 = CYP2D6 inhibitor (drug-drug interaction risk).
CYP2C9 InhibitionclaTDCAUC=0.904
F1 0.75
AUC=0.893
F1 0.65
1 = CYP2C9 inhibitor (drug-drug interaction risk).
Ames MutagenicityclaTDCAUC=0.895
F1 0.84
AUC=0.768
F1 0.60
1 = mutagenic in the Ames test. Lower is safer.
DILI (Hepatotoxicity)claTDCAUC=0.871
F1 0.81
AUC=0.582
F1 0.33
1 = drug-induced liver injury. Lower is safer.
Hepatocyte ClearanceregTDCR²=0.240
RMSE 0.492
R²=0.077
RMSE 0.540
Intrinsic hepatocyte clearance (log10-trained). Lower = more stable.
Plasma Protein BindingregTDCR²=0.490
RMSE 13.230
R²=0.456
RMSE 12.539
Fraction bound to plasma protein (%). Very high (>99%) limits free drug.
Acute Toxicity (LD50)regTDCR²=0.604
RMSE 0.581
R²=0.318
RMSE 0.888
Rat oral LD50 as -log(mol/kg). Higher = more toxic.

Random split = deterministic 80/20 (MoleculeNet baseline). Scaffold split = Bemis-Murcko scaffolds held out entirely — the harder, more realistic generalization estimate, and the number you should trust. Regression metrics for log-transformed PK endpoints are reported in log space.

Screened library — 30 molecules (from 30 input rows), ranked by developability score

MoleculeScoreAqueous Solubility (logS)
log(mol/L)
Lipophilicity (logD 7.4)
logD
BBB Penetration
P(penetrant)
hERG Cardiotox (blocker)
P(blocker)
Caco-2 Permeability
log(Papp cm/s)
CYP3A4 Inhibition
P(inhibitor)
CYP2D6 Inhibition
P(inhibitor)
CYP2C9 Inhibition
P(inhibitor)
Ames Mutagenicity
P(mutagen)
DILI (Hepatotoxicity)
P(hepatotox)
Hepatocyte Clearance
uL/min/1e6
Plasma Protein Binding
% bound
Acute Toxicity (LD50)
-log(mol/kg)
Drug-likeness & alerts
Salbutamol
CC(C)(C)NCC(O)c1ccc(O)c(CO)c1
87-1.59 0.423%4%-5.452%15%0%10%1%12.9221.762.42Ro5 0v Veber Ghose QED 0.639 SA 2.74
clean
Ibuprofen
CC(C)Cc1ccc(C(C)C(=O)O)cc1
81-3.170.9142%41%-4.461%3%2%2%45%19.0275.522.41Ro5 0v Veber Ghose QED 0.822 SA 2.19
clean
Paracetamol
CC(=O)Nc1ccc(O)cc1
80-1.190.4798%4%-4.571%3%7%25%25%27.2330.891.84Ro5 0v Veber Ghose QED 0.595 SA 1.41
Brenk:hydroquinone
Morphine
CN1CCC23C(O)=CCC1C2Cc1ccc(O)c2c1C3O2
79-3.35 1.32100%38%-4.855%34% 4% 35%5%46.51 64.703.31Ro5 0v Veber Ghose QED 0.769 SA 5.22
clean
Caffeine
Cn1c(=O)c2c(ncn2C)n(C)c1=O
76-0.980.0798%48%-4.370%0%1%17%71%4.6527.962.43Ro5 0v Veber Ghose QED 0.538 SA 2.3
clean
Aspirin
CC(=O)Oc1ccccc1C(=O)O
76-1.68-0.1849%3%-4.580%0%4%5%66%16.4051.132.22Ro5 0v Veber Ghose QED 0.55 SA 1.58
Brenk:phenol_ester
Propranolol
CC(C)NCC(O)COc1cccc2ccccc12
76-3.400.9148%88%-4.768%48%2%2%4%89.2786.102.33Ro5 0v Veber Ghose QED 0.838 SA 2.3
clean
Hydroquinone
Oc1ccc(O)cc1
720.100.8469%2%-4.773%5%13%22%39%31.4636.382.27Ro5 0v Veber Ghose QED 0.491 SA 1.53
Brenk:hydroquinone NIH:para_hydroquinone
Thalidomide
O=C1CCC(N2C(=O)c3ccccc3C2=O)C(=O)N1
70-2.740.68 99%4%-4.661%1%1%37%95%8.8780.342.71Ro5 0v Veber Ghose QED 0.723 SA 2.6
Brenk:phthalimide
Penicillin_G
CC1(C)SC2C(NC(=O)Cc3ccccc3)C(=O)N2C1…
70-3.02 0.282%33%-5.691%3%1%3%45%17.5959.192.33Ro5 0v Veber Ghose QED 0.798 SA 3.2
NIH:betalactam
Diazepam
CN1C(=O)CN=C(c2ccccc2)c2ccc(Cl)cc21
69-3.802.88100%43%-4.3933%14%66%3%96%6.2191.082.31Ro5 0v Veber Ghose QED 0.792 SA 2.16
clean
Lovastatin
CCC(C)C(=O)OC1CC(C)C=C2C=CC(C)C(CCC3…
68-5.613.0398%66%-5.0275%1%5%12% 16%28.4287.292.55Ro5 0v Veber Ghose QED 0.672 SA 4.69
clean
Fluoxetine
CNCCC(Oc1ccc(C(F)(F)F)cc1)c1ccccc1
66-5.451.8098%99%-4.6145%64%45%11% 3%58.1092.712.48Ro5 0v Veber Ghose QED 0.852 SA 2.45
clean
Warfarin
CC(=O)CC(c1ccccc1)c1c(O)c2ccccc2oc1=…
66-4.232.0150%36%-4.8219%6%80%4% 89%3.1995.853.55Ro5 0v Veber Ghose QED 0.748 SA 2.63
Brenk:cumarine
Metformin
CN(C)C(=N)N=C(N)N
63-0.47 -0.42 65% 2% -6.15 0%1%1%53% 2%4.33 49.52 2.27 Ro5 0v Veber Ghose QED 0.282 SA 3.21
Brenk:imine_1 Brenk:imine_2
Terfenadine
CC(C)(C)c1ccc(C(O)CCCN2CCC(C(O)(c3cc…
59-6.433.6313%98%-5.453%86%3%2% 2%53.7395.352.32Ro5 1v Veber Ghose QED 0.397 SA 2.86
clean
Sildenafil
CCCc1nn(C)c2c(=O)n(CC)c(-c3cc(S(=O)(…
57-4.19 1.9493% 89%-4.8951%2%16% 31% 97%21.9273.212.49 Ro5 1v Veber Ghose QED 0.465 SA 2.75
clean
Curcumin
COc1cc(/C=C/C(=O)CC(=O)/C=C/c2ccc(O)…
53-4.612.0869%65%-4.8553%4%66%16%86%27.0797.472.37Ro5 0v Veber Ghose QED 0.548 SA 2.43
Brenk:beta-keto/anhydride Brenk:Michael_acceptor_1
Toluidine_blue
CN(C)c1ccc2nc3ccc(=[N+](C)C)cc-3sc2c…
52-4.01 1.51100% 71%-4.47 21% 18% 4% 56% 81%33.7684.692.51 Ro5 0v Veber Ghose QED 0.505 SA 2.74
Brenk:Polycyclic_aromatic_hydrocarbon_2 Brenk:quaternary_nitrogen_1
Verapamil
COc1ccc(CCN(C)CCCC(C#N)(c2ccc(OC)c(O…
50-6.52 2.3768%96%-4.7889%25%21%25% 68%25.7287.123.34Ro5 1v Veber Ghose QED 0.42 SA 2.94
clean
Nitrofuran_ex
O=[N+]([O-])c1ccc(/C=N/N2CCOCC2)o1
50-2.580.1999% 2%-4.562%1%2%99%96%9.7042.63 2.39Ro5 0v Veber Ghose QED 0.433 SA 2.74
Brenk:imine_1 Brenk:nitro_group Brenk:Oxygen-nitrogen_single_bond
Cisapride
COc1cc(C(=O)NC2CCN(CCCOc3ccc(F)cc3)C…
50-4.18 2.3199%94%-4.8742%47%21%20% 77%15.6985.412.91Ro5 0v Veber Ghose QED 0.469 SA 2.98
Brenk:Aliphatic_long_chain Brenk:aniline
Astemizole
COc1ccc(CCN2CCC(Nc3nc4ccccc4n3Cc3ccc…
50-6.61 3.5745%99%-5.1332%94%35%49%17%35.8095.122.79Ro5 1v Veber Ghose QED 0.386 SA 2.19
clean
Imatinib
Cc1ccc(NC(=O)c2ccc(CN3CCN(C)CC3)cc2)…
50-5.83 2.6331%95%-5.3882%22%54%4% 98%19.5593.082.76 Ro5 0v Veber Ghose QED 0.389 SA 2.33
clean
Atorvastatin
CC(C)c1c(C(=O)Nc2ccccc2)c(-c2ccccc2)…
47-6.27 2.7124%89%-5.4723%4%64%3%87%14.2999.872.55 Ro5 2v Veber Ghose QED 0.163 SA 3.31
clean
Rhodanine_frag
O=C1CSC(=S)N1
36-1.580.18 100% 12% -4.58 1% 0% 5% 39%61% 9.45 46.13 2.88Ro5 0v Veber Ghose QED 0.476 SA 3.46
PAINS:rhod_sat_A(33) Brenk:Thiocarbonyl_group
Caffeic_acid
O=C(O)/C=C/c1ccc(O)c(O)c1
35-1.780.6059%1% -5.343%1%6%17%22%9.8951.241.95Ro5 0v Veber Ghose QED 0.472 SA 2.04
PAINS:catechol_A(92) Brenk:catechol Brenk:Michael_acceptor_1
Loratadine
CCOC(=O)N1CCC(=C2c3ccccc3CCc3cc(Cl)c…
28-6.493.6693%98%-4.9138%25%76%12% 34%19.7395.803.40Ro5 1v Veber Ghose QED 0.653 SA 2.36
PAINS:styrene_A(13)
Quercetin
O=c1c(O)c(-c2ccc(O)c(O)c2)oc2cc(O)cc…
26-3.302.1729%13%-5.9862%12%9%57%96%10.7492.49 2.58Ro5 0v Veber Ghose QED 0.434 SA 2.54
PAINS:catechol_A(92) Brenk:catechol NIH:gte_5_phenolic_OH
Ketoconazole
CC(=O)N1CCN(c2ccc(OCC3COC(Cn4ccnc4)(…
22-6.053.3840%99%-4.8998%24%87%25% 99%10.0795.723.16Ro5 1v Veber Ghose QED 0.455 SA 3.44
PAINS:anil_di_alk_C(246)

Each prediction carries a confidence dot (High Medium Low, from ensemble spread + applicability domain). A marks an out-of-domain prediction (nearest training neighbour below the Tanimoto threshold — extrapolation, treat with caution). Classification cells show predicted probability; orange = ≥50%. For safety endpoints (hERG, CYP, Ames, DILI, LD50) higher = worse and is penalized in the score. PAINS / Brenk / NIH structural alerts are listed with the matched catalog name; hover an alert chip for the matched substructure SMILES.