enumerate a large amount of rdkit valid SMILES.
Sometimes, the doRandom=True flag of Chem.MolToSmiles() in rdkit just doesn't cut it. For example, the topical nasal decongestant c1ccccc1CC(C)NC has merely 40 unique random SMILES. Making use of the "use ring operators as normal bonds" trick, we can do better than that and generate a large amount of ugly but rdkit-valid SMILES! Because using this trick makes it possible to use any atom ordering, this means a total of N! SMILES of this type are possible, with N the amount of atoms and ! the factorial (I'm not shouting, I'm calm). In case there's symmetry the number will be a bit lower.
The notion of generating SMILES for any ordering was previously explored, implemented and blogged about by Andrew Dalke in 2010 in a blog post titled Reordering SMILES atom order.
should work in any python env with reasonably up to date rdkit.
python ultra_enumerate.py --smiles "c1ccccc1CC(C)NC" -o output.smi -N 100000
with -N the amount of SMILES you want, -o the output (if left empty it will print to stdout) and --smiles the SMILES you want to have variants of.
Note that they will look like this.
c:1:2.C345.c:6:7.C4.c:8:1.C93.N5%10.c:%11:29.C%10.c:6:%11.c:7:8
C1.C2.c:3:4.c:5:6.c:6:7.C819.c:3:%10.c:4:5.c:%10:7%11.N92.C%118
C123.c:4:5.c:6:7.C2.C8.c:9:%10%11.N38.c:4:9.c:5:6.C%111.c:7:%10
c:1:2.c:3:4.C567.c:8:49.C6.c:2:%10.c:1:8.C95.C%11.N7%11.c:%10:3
c:1:2.c:3:4.N56.c:7:1.c:4:28.c:9:7.C6.c:3:9.C%10.C8%11.C%11%105
c:1:2.c:3:4.c:4:56.c:7:5.c:2:7.C89%10.N%10%11.C9.c:3:1.C68.C%11
C123.c:4:5.C2.c:6:7.C8.c:5:9%10.c:4:6.N38.c:7:%11.c:%11:9.C%101
c:1:2.C3.c:4:5.c:1:6.c:6:4.C789.C%107.c:2:%11%10.C8.N93.c:5:%11
c:1:23.c:4:5.c:6:7.c:7:2.c:4:1.C8.c:5:6.C39.N%10%11.C%11.C98%10
c:1:23.N45.c:6:7.C894.c:7:2.c:%10:%11.C5.C38.c:%10:1.c:%11:6.C9
The speed is relatively practical, for example the output 100000 nasal decongestant random SMILES take about 2 seconds to generate.
For the moment stereochemistry is ignored.