Skip to content

Reference recognizer, chembl_test_macfonts

Failure breakdown

Attempted: 6000 | exact: 3000 (0.500)

failure mode n share of all share of failures what it points at
unparseable output 0 0.000 0.000 decoder emitted a string RDKit cannot read
stereochemistry only 0 0.000 0.000 skeleton correct; wedge/hash reading wrong
right formula, wrong bonds 0 0.000 0.000 atoms counted correctly, connectivity misread
other structural error 3000 0.500 1.000 a genuinely different molecule

Accuracy by molecule size

heavy atoms n exact valid
<=10 18 0.500 1.000
<=15 168 0.500 1.000
<=20 648 0.500 1.000
<=25 1482 0.500 1.000
<=30 1482 0.500 1.000
<=35 1152 0.500 1.000
<=40 654 0.500 1.000
>40 396 0.500 1.000

Accuracy by rendering style

style n exact valid
clean 2000 0.500 1.000
degraded 2000 0.500 1.000
varied 2000 0.500 1.000

Sequence length

  • Reference SMILES length, all attempted: median 47 tokens
  • Reference SMILES length, correct predictions: median 47 tokens

Longest failures

reference predicted
O=S(=O)(O)c1ccc(-c2ccc(S(=O)(=O)Nc3cc(S(=O)(=O)O)cc4cc(S(=O)(=O)O)cc(S CC[C@@H](c1ccc(O)cc1)[C@H](CCCCn1cc(C(=O)O)c(=O)[nH]1)c1ccc(O)cc1
O=S(=O)(O)c1ccc(-c2ccc(S(=O)(=O)Nc3cc(S(=O)(=O)O)cc4cc(S(=O)(=O)O)cc(S OC[C@@H]1CCN(c2cc(-c3ccccc3)nc3ccccc23)C[C@@H]1O
O=S(=O)(O)c1ccc(-c2ccc(S(=O)(=O)Nc3cc(S(=O)(=O)O)cc4cc(S(=O)(=O)O)cc(S O=C(C[C@@H](CCCC1CCCCC1)c1nc(CNS(=O)(=O)c2cccnc2)no1)NO
CC(=O)OC[C@H]1O[C@@H](O[C@H]2[C@H](OC(C)=O)[C@@H](OC(C)=O)[C@H](S(N)(= CCOC(=O)c1ccccc1N[C@@H]1c2cc3c(cc2[C@@H](C2=CC(=O)C(=O)C(OC)=C2)[C@H]2
CC(=O)OC[C@H]1O[C@@H](O[C@H]2[C@H](OC(C)=O)[C@@H](OC(C)=O)[C@H](S(N)(= O=S(=O)(O)c1cc(Cl)c(O)c2ncc(Cl)cc12