Wrong Predictions Browser

Every incorrect prediction, fully disclosed. We believe transparency about failure modes is essential to scientific credibility.

wrong predictions

96.54%

validated accuracy

1,533

compounds tested

100%

fully disclosed

Errors only

Showing 50 compound classes with 93 total wrong predictions

Compound Class↑↓	Scope	N↑↓	Wrong↓	LOO Accuracy↑↓	Method	Source
4-AQ (other) Pharmaceutical	Exploratory	110	34	69.1%	v2	ChEMBL 36
tert-Carbinol triazole Pharmaceutical	Validated	170	14	91.8%	v6	ChEMBL 36
Aryl-ethanolamines Pharmaceutical	Validated	181	11	93.9%	v1	ChEMBL 36
Amino-alcohols (REDACTED) Pharmaceutical	Validated	92	9	90.2%	v8	ChEMBL 36
Morphinan antagonists Pharmaceutical	Validated	259	8	96.9%	v2	ChEMBL 36
Catecholamines (3,4-diol) Pharmaceutical	Validated	61	6	90.2%	v1	ChEMBL 36
Classical 4-AQ (7-Cl) Pharmaceutical	Exploratory	48	5	89.6%	v2	ChEMBL 36
Pyrrolidinone SV2A Pharmaceutical	Validated	79	2	97.5%	v1	ChEMBL 36
8-Aminoquinolines Pharmaceutical	Validated	22	2	90.9%	v1	ChEMBL 36
Arylpropionic acids Pharmaceutical	Validated	94	1	98.9%	v1	ChEMBL 36
N-methyl pipecoloxylidides Pharmaceutical	Exploratory	1	1	0.0%	v1	Legacy
β-blockers (N-tBu) Pharmaceutical	Validated	22	0	100.0%	v1	ChEMBL 36
Morphinan agonists Pharmaceutical	Validated	18	0	100.0%	v1	Legacy
ACCase herbicides Agricultural	Validated	17	0	100.0%	v1	Legacy
Pipecoloxylidides (N≥C3) Pharmaceutical	Validated	10	0	100.0%	v1	Legacy
DHP (ester-sidechain) Pharmaceutical	Validated	16	0	100.0%	v1	ChEMBL 36
DHP (C4 ring-chiral) Pharmaceutical	Validated	43	0	100.0%	v4	ChEMBL 36
Benzamide antipsychotics Pharmaceutical	Validated	306	0	100.0%	v2	ChEMBL 36
ACE inhibitors Pharmaceutical	Validated	23	0	100.0%	v1	ChEMBL 36
Proton pump inhibitors Pharmaceutical	Validated	16	0	100.0%	v1	ChEMBL 36
Axial atropisomers Pharmaceutical	Validated	21	0	100.0%	v7	ChEMBL 36
Triazole dioxolanes Agricultural	Exploratory	7	0	100.0%	v6	Legacy
Triazole open-chain Agricultural	Validated	7	0	100.0%	v6	Legacy
Triazole fused-ring Agricultural	Exploratory	6	0	100.0%	v6	Legacy
α-Carbon triazoles Agricultural	Exploratory	6	0	100.0%	v1	Legacy
Metalaxyl (MeO-acetyl) Agricultural	Exploratory	4	0	100.0%	v1	Legacy
Benalaxyl (Ph-acetyl) Agricultural	Exploratory	4	0	100.0%	v1	Legacy
Coumarin anticoagulants Pharmaceutical	Validated	16	0	100.0%	v1	Legacy
PARP inhibitors Pharmaceutical	Exploratory	16	0	100.0%	v1	Legacy
ALK inhibitors Pharmaceutical	Exploratory	11	0	100.0%	v1	Legacy
SNRI antidepressants Pharmaceutical	Exploratory	19	0	100.0%	v1	Legacy
Diamide insecticides Agricultural	Exploratory	4	0	100.0%	v1	Legacy
Neonicotinoids Agricultural	Exploratory	4	0	100.0%	v1	Legacy
Amino acid sulfoxides Pharmaceutical	Validated	8	0	100.0%	v1	Legacy
Pyrimidine nucleosides Pharmaceutical	Validated	18	0	100.0%	v1	Legacy
Oxazolidinone antibiotics Pharmaceutical	Exploratory	10	0	100.0%	v1	Legacy
Fluoroquinolones Pharmaceutical	Exploratory	8	0	100.0%	v1	Legacy
Propylamine antihistamines Pharmaceutical	Exploratory	16	0	100.0%	v1	Legacy
Arylcyclohexylamines Pharmaceutical	Exploratory	11	0	100.0%	v1	Legacy
Thienopyridines Pharmaceutical	Exploratory	3	0	100.0%	v1	Legacy
Chloroacetanilides Agricultural	Exploratory	4	0	100.0%	v1	Legacy
Imidazolinones Agricultural	Exploratory	4	0	100.0%	v1	Legacy
3-OH Benzodiazepines Pharmaceutical	Exploratory	6	0	100.0%	v1	Legacy
FOP herbicides Agricultural	Exploratory	4	0	100.0%	v1	Legacy
Pyrethroid insecticides Agricultural	Exploratory	9	0	100.0%	v1	Legacy
Benzothiazepine CCBs Pharmaceutical	Exploratory	2	0	100.0%	v1	Legacy
Carbapenem antibiotics Pharmaceutical	Validated	8	0	100.0%	v1	Legacy
Cephalosporin antibiotics Pharmaceutical	Validated	10	0	100.0%	v1	Legacy
PI3K inhibitors Pharmaceutical	Validated	16	0	100.0%	pending-subclassification	Legacy
JAK pyrrolopyrimidines Pharmaceutical	Exploratory	24	0	100.0%	pending-subclassification	Legacy

Error distribution (validated compound classes)

tert-Carbinol triazole

14 (26.4%)

Aryl-ethanolamines

11 (20.8%)

Amino-alcohols (REDACTED)

9 (17.0%)

Morphinan antagonists

8 (15.1%)

Catecholamines (3,4-diol)

6 (11.3%)

Pyrrolidinone SV2A

2 (3.8%)

8-Aminoquinolines

2 (3.8%)

Arylpropionic acids

1 (1.9%)

How wrong predictions are identified

Wrong predictions are identified through leave-one-out (LOO) cross-validation against ChEMBL release 36. For each compound in a compound class, we remove it from the training set, predict its chirality using the compound class's computational method, and compare to the known CIP label parsed from the compound's SMILES via RDKit. A prediction is “wrong” when the predicted active enantiomer (R or S) does not match the experimentally determined assignment.

Compound classes without ChEMBL cross-validation data (shown as “Legacy only”) use prospective blind validation instead: compounds were selected and locked before predictions were run.

Read the full methodology →