Sorry, looks like my baby is getting old ... :-) Markus
On Wed, Feb 25, 2015 at 7:26 AM, Greg Landrum <greg.land...@gmail.com> wrote: > To close the loop here: after an email exchange with Marc Nicklaus and Wolf > Ihlenfeldt, it looks like the problem is that the NCI website is using an > older version of the CACTVS toolkit to do the SMILES->InChI conversion. That > older version contains a bug that has since been fixed. Marc is now aware of > the problem. > > The RDKit was, at least in this case, not responsible for the bad InChIs. > <whew!> :-) > > Best, > -greg > > > > > On Tue, Feb 24, 2015 at 8:27 AM, Greg Landrum <greg.land...@gmail.com> > wrote: >> >> >> The InChIs have me confused. >> >> I'm going to simplify the below by just showing the input SMILES, the >> current (=master) RDKit InChI and the PubChem InChI >> >> On Mon, Feb 23, 2015 at 10:54 AM, JP <jeanpaul.ebe...@inhibox.com> wrote: >>> >>> >>> Here is the list (first inchi is the 2014_09_2, second one is the >>> 2015.03.1pre generated one, third inchi is the cactus.nci.nih.gov): >>> >>> O=C(/N=c1/[nH]ncs1)[C@H]1CC[C@H](Cn2cnc3ccccc3c2=O)CC1 >>> >>> InChI=1S/C18H19N5O2S/c24-16(21-18-22-20-11-26-18)13-7-5-12(6-8-13)9-23-10-19-15-4-2-1-3-14(15)17(23)25/h1-4,10-13H,5-9H2,(H,21,22,24)/t12-,13- >>> # RDKit 2015.03.1pre >>> >>> InChI=1S/C18H29N5O2S/c24-16(21-18-22-20-11-26-18)13-7-5-12(6-8-13)9-23-10-19-15-4-2-1-3-14(15)17(23)25/h12-15,19-20H,1-11H2,(H,21,22,24)/t12-,13-,14?,15? >>> # cactus.nci.nih.gov >>> >>> O=C(/N=c1\[nH]c(-c2ccccn2)cs1)[C@H]1CC[C@H](Cn2cnc3ccccc3c2=O)CC1 >>> InChI=1S/C24H23N5O2S/c30-22(28-24-27-21(14-32-24)20-7-3-4-12-25-20)17-10-8-16(9-11-17)13-29-15-26-19-6-2-1-5-18(19)23(29)31/h1-7,12,14-17H,8-11,13H2,(H,27,28,30)/t16-,17- >>> >>> InChI=1S/C24H39N5O2S/c30-22(28-24-27-21(14-32-24)20-7-3-4-12-25-20)17-10-8-16(9-11-17)13-29-15-26-19-6-2-1-5-18(19)23(29)31/h16-21,25-26H,1-15H2,(H,27,28,30)/t16-,17-,18?,19?,20?,21? >>> >>> CCOC(=O)Cc1cs/c(=N/C(=O)[C@H]2CC[C@H](Cn3cnc4ccccc4c3=O)CC2)[nH]1 >>> InChI=1S/C23H26N4O4S/c1-2-31-20(28)11-17-13-32-23(25-17)26-21(29)16-9-7-15(8-10-16)12-27-14-24-19-6-4-3-5-18(19)22(27)30/h3-6,13-16H,2,7-12H2,1H3,(H,25,26,29)/t15-,16- >>> >>> InChI=1S/C23H36N4O4S/c1-2-31-20(28)11-17-13-32-23(25-17)26-21(29)16-9-7-15(8-10-16)12-27-14-24-19-6-4-3-5-18(19)22(27)30/h15-19,24H,2-14H2,1H3,(H,25,26,29)/t15-,16-,17?,18?,19? >>> >>> COCc1n[nH]/c(=N/C(=O)[C@H]2CC[C@H](Cn3cnc4ccccc4c3=O)CC2)s1 >>> InChI=1S/C20H23N5O3S/c1-28-11-17-23-24-20(29-17)22-18(26)14-8-6-13(7-9-14)10-25-12-21-16-5-3-2-4-15(16)19(25)27/h2-5,12-14H,6-11H2,1H3,(H,22,24,26)/t13-,14- >>> >>> InChI=1S/C20H33N5O3S/c1-28-11-17-23-24-20(29-17)22-18(26)14-8-6-13(7-9-14)10-25-12-21-16-5-3-2-4-15(16)19(25)27/h13-17,21,23H,2-12H2,1H3,(H,22,24,26)/t13-,14-,15?,16?,17? >>> >>> COC(=O)c1[nH]/c(=N\C(=O)[C@H]2CC[C@H](Cn3cnc4ccccc4c3=O)CC2)sc1C(C)C >>> InChI=1S/C24H28N4O4S/c1-14(2)20-19(23(31)32-3)26-24(33-20)27-21(29)16-10-8-15(9-11-16)12-28-13-25-18-7-5-4-6-17(18)22(28)30/h4-7,13-16H,8-12H2,1-3H3,(H,26,27,29)/t15-,16- >>> >>> InChI=1S/C24H38N4O4S/c1-14(2)20-19(23(31)32-3)26-24(33-20)27-21(29)16-10-8-15(9-11-16)12-28-13-25-18-7-5-4-6-17(18)22(28)30/h14-20,25H,4-13H2,1-3H3,(H,26,27,29)/t15-,16-,17?,18?,19?,20? >>> >>> CC(C)[C@H]1CC[C@H](C(=O)N[C@H](Cc2ccccc2)C(=O)/N=c2\[nH]ncs2)CC1 >>> InChI=1S/C21H28N4O2S/c1-14(2)16-8-10-17(11-9-16)19(26)23-18(12-15-6-4-3-5-7-15)20(27)24-21-25-22-13-28-21/h3-7,13-14,16-18H,8-12H2,1-2H3,(H,23,26)(H,24,25,27)/t16-,17-,18-/m1/s1 >>> >>> InChI=1S/C21H36N4O2S/c1-14(2)16-8-10-17(11-9-16)19(26)23-18(12-15-6-4-3-5-7-15)20(27)24-21-25-22-13-28-21/h14-18,22H,3-13H2,1-2H3,(H,23,26)(H,24,25,27)/t16-,17-,18-/m1/s1 >> >> >> If you look in the formula layer for the InChIs from PubChem, you will see >> that they all have *way* too many H atoms. I think there's something about >> the structures that is confusing the pubchem/cactvs conversion code. >> >> Compare these two outputs. >> Aromatic form: >> >> http://cactus.nci.nih.gov/chemical/structure/O=C(N=c1[nH]ncs1)C1CCC(Cn2cnc3ccccc3c2=O)CC1/stdinchi >> produces: >> >> InChI=1S/C18H29N5O2S/c24-16(21-18-22-20-11-26-18)13-7-5-12(6-8-13)9-23-10-19-15-4-2-1-3-14(15)17(23)25/h12-15,19-20H,1-11H2,(H,21,22,24) >> >> Kekule form: >> >> http://cactus.nci.nih.gov/chemical/structure/O=C(/N=C1/[NH]N=CS1)[C@H]1CC[C@H](CN2C=NC3=CC=CC=C3C2=O)CC1/stdinchi >> produces: >> >> InChI=1S/C18H19N5O2S/c24-16(21-18-22-20-11-26-18)13-7-5-12(6-8-13)9-23-10-19-15-4-2-1-3-14(15)17(23)25/h1-4,10-13H,5-9H2,(H,21,22,24)/t12-,13- >> >> In fact, converting the 5 membered ring to kekule form is enough: >> >> http://cactus.nci.nih.gov/chemical/structure/O=C(N=C1[NH]N=CS1)C1CCC(Cn2cnc3ccccc3c2=O)CC1/stdinchi >> produces: >> >> InChI=1S/C18H19N5O2S/c24-16(21-18-22-20-11-26-18)13-7-5-12(6-8-13)9-23-10-19-15-4-2-1-3-14(15)17(23)25/h1-4,10-13H,5-9H2,(H,21,22,24) >> >> This can't be true. >> >> We can further simplify things to track down the problem: >> >> http://cactus.nci.nih.gov/chemical/structure/N=c1[nH]ncs1/stdinchi >> InChI=1S/C2H5N3S/c3-2-5-4-1-6-2/h4H,1H2,(H2,3,5) >> >> vs >> >> http://cactus.nci.nih.gov/chemical/structure/O=c1[nH]ncs1/stdinchi >> InChI=1S/C2H2N2OS/c5-2-4-3-1-6-2/h1H,(H,4,5) >> >> It seems to be the exocyclic bond to an atom with Hs. This is ok: >> http://cactus.nci.nih.gov/chemical/structure/O=c1occo1/stdinchi >> InChI=1S/C3H2O3/c4-3-5-1-2-6-3/h1-2H >> >> but both of these are wrong: >> http://cactus.nci.nih.gov/chemical/structure/N=c1occo1/stdinchi >> InChI=1S/C3H5NO2/c4-3-5-1-2-6-3/h4H,1-2H2 >> >> http://cactus.nci.nih.gov/chemical/structure/C=c1occo1/stdinchi >> InChI=1S/C4H6O2/c1-4-5-2-3-6-4/h1-3H2 >> >> I'm pretty sure that this is not the RDKit doing the wrong thing. >> >> @Markus: what would be the best way to report this to the NCI CADD guys? >> >> -greg >> > ------------------------------------------------------------------------------ Dive into the World of Parallel Programming The Go Parallel Website, sponsored by Intel and developed in partnership with Slashdot Media, is your hub for all things parallel software development, from weekly thought leadership blogs to news, videos, case studies, tutorials and more. Take a look and join the conversation now. http://goparallel.sourceforge.net/ _______________________________________________ Rdkit-discuss mailing list Rdkit-discuss@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/rdkit-discuss
