Skip to main content

Table 3 Results on the GuacaMol benchmark

From: EvoMol: a flexible and interpretable evolutionary algorithm for unbiased de novo molecular generation

Benchmark SOTA methods EvoMol
SMILES LSTM Graph GA CReM [5] MSO
[9]
Primary actions All actions From Methane Best run Best scores
Celecoxib rediscovery 1.000 1.000 1.000 1.000 0.714 0.978 0.923 1.000 1.000
Troglitazone rediscovery 1.000 1.000 1.000 1.000 0.936 1.000 0.676 1.000 1.000
Thiotixene rediscovery 1.000 1.000 1.000 1.000 0.852 0.876 0.695 1.000 1.000
Aripiprazole similarity 1.000 1.000 1.000 1.000 1.000 1.000 0.964 1.000 1.000
Albuterol similarity 1.000 1.000 1.000 1.000 1.000 1.000 0.878 1.000 1.000
Mestranol similarity 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000
\(\hbox {C}_{11}\hbox {H}_{24}\) 0.993 0.971 0.966 0.997 1.000 1.000 1.000 1.000 1.000
\(\hbox {C}_9\hbox {H}_{10}\hbox {N}_2\hbox {O}_2\hbox {PF}_2\hbox {Cl}\) 0.879 0.982 0.940 1.000 1.000 0.998 1.000 1.000 1.000
Median molecules 1 0.438 0.406 0.371 0.437 0.446 0.455 0.455 0.455 0.455
Median molecules 2 0.422 0.432 0.434 0.395 0.411 0.417 0.286 0.417 0.417
Osimertinib MPO 0.907 0.953 0.995 0.966 0.959 0.955 0.911 0.969 0.978
Fexonadine MPO 0.959 0.998 1.000 1.000 0.966 1.000 0.981 1.000 1.000
Ranolazine MPO 0.855 0.920 0.969 0.931 0.943 0.966 0.967 0.957 1.000
Perindopril MPO 0.808 0.792 0.815 0.834 0.809 0.845 0.789 0.827 0.884
Amlodipine MPO 0.894 0.894 0.902 0.900 0.874 0.867 0.796 0.869 0.906
Sitagliptin MPO 0.545 0.891 0.763 0.868 0.943 0.915 0.946 0.926 0.966
Zaleplon MPO 0.669 0.754 0.770 0.764 0.791 0.791 0.771 0.793 0.810
Valsartan SMARTS 0.978 0.990 0.994 0.994 0.999 0.998 0.000 0.998 1.000
deco hop 0.996 1.000 1.000 1.000 1.000 1.000 0.607 1.000 1.000
scaffold hop 0.998 1.000 1.000 1.000 0.989 1.000 0.655 1.000 1.000
total 17.340 17.983 17.919 18.086 17.632 18.060 15.298 18.210 18.415
total MPO only 5.637 6.202 6.214 6.263 6.286 6.339 6.160 6.341 6.544
  1. SMILES LSTM and Graph GA values are reported form the GuacaMol article. Values for CReM and MSO methods are reported from their respective articles. Values in the three first EvoMol columns correspond to the mean scores on 10 executions for different initial conditions and parameters. The best run column reports the values of the best execution. The best scores corresponds for each task to the best encountered value during all executions independently of the run and thus, the totals in this column are virtual