some kind of blog

I’ve been playing with InChIs these days quite a bit and have been quite suprised by the results that I got. The one thing that is very surprising indeed, is that the InChI - and therefore the InChIKey, should one want to use these to index a chemical compound database - one gets, seems to depend on the input format for the InChI generator.

After having seen some more InChI related things atthe BioIT World yesterday, I thought I’d have another go at some InChI stuff. And I was again quite surprised to what I found, and actually even more confused.

Here’s what I did:

I took two molecules from PubChem (diaisostereomers) cis- and trans-1,2-dihydroxy-cyclohexane (1,2-cyclohexanediol). I created InChIKeys for both of them, using different inputs. And not at all did I end up with two different InChIs for these two structures that were the same.

The SMILES and InChIKey from PubChem are the following:
cis-1,2-cyclohexanediol: C1CC[C@@H]([C@@H](C1)O)O - PFURGBBHAOXLIO-OLQVQODUSA-N
trans-1,2-cyclohexanediol: C1CC[C@H]([C@@H](C1)O)O - PFURGBBHAOXLIO-PHDIDXHHSA-N

I used the Standard InChI generator v1.2 released in January 2009 by the IUPAC, more specifically the Linux binary. Because this software doesn’t read SMILES, one has to convert the SMILES input molecules to SD files. This was done with OpenBabel and Pipeline Pilot. In addition, I used CML obtained from OpenBabel as input, too.

And here is what I obtained:

Now, this raises the following questions:

1. First of all, why do I not just get the same InChI for everything? Isn’t it supposed to canonicalize the structure to a fair extent (I know it does) and come up with the same InChI?
2. If there are different ways of getting a presumably valid InChI for a compound, which one should one take?
3. Why do I get the same InChI for different stereoisomers when the 3D coordinates come from OpenBabel, but not if the coordinates come from Pipeline Pilot? Is the 3D coordinate generation in OpenBabel the problem?
4. Why do I get the correst stereoinformation in the InChIKey when I use CML without any coordinates?
5. Why do I get a different hash for the connectivity part when I use CML without any coordinates? FWITZFBVZWAIRX instead of PFURGBBHAOXLIO?
6. And then, if CML without 3D coordinates gives the right stereochemistry in the InChI, why doesn’t 3D CML with coordinates from OpenBabel?
7. 3D CML from PipelinePilot coordinates also doesn’t get the stereochemistry part right; instead, there’s yet a different connectivity part DQQAEJAQEMHJBB.
8. Why should the stereochemistry come out right when 2D coordinates are used, but not if 0D coordinates are used?
9. Why not get InChIs directly from SMILES?

Any comments on this anybody? I seriously wonder.

There are actually a few bugs in this version. Shortly I’ll put up a new version here.

Couldn’t find a program to get all ring systems out of an SDF file, along with how often they occur. So I wrote a script in python that does exactly that job. It gets the smallest set of smallest rings (SSSR) from pybel: once a molecule is read in (e.g., mol=pybel.readstring(”smi”, smiles)), then you have the SSSR in mol.sssr, which is a vector of OBRing objects (see the OpenBabel documentation for more info about that).

You can iterate over this vector in a standard pythonic fashion, e.g., for ring in mol.sssr: pass. The ring size is easily accessed by ring.PathSize(), the atoms in the ring are stored in the member variable _path, e.g., ring._path will give you the atoms in the ring.

The script checks for fused ring systems by identifying any shared atoms between any members of the SSSR. This is achieved by intersection of the sets of member atoms of any two ring systems. Two rings are considered to be a ring system if they share at least one atom, i.e., strictly speaking it is not fused but rather a spiro system. This behaviour can be changed by changing if len(intsec) in function GetFusedRingsMatrix(mol) to if len(intsec) > 1.

Should you want to get all individual elements of the SSSR, instead of the fused/linked rings as one ring system, then it should suffice to supply a manually crafted matrix as the one returned by GetFusedRingsMatrix(). That would be something like l=len(mol.sssr) and FusedRingsMatrix = [[0 for x in range(l)] for y in range(l)]. Then all rings are supposed to be unconnected.

The script also includes exocyclic double bonds as part of rings they may be linked to.

All fragments are written out to a file fragments.sdf, and the number how often that fragment was encountered in the structure file supplied is written into the SDF field COUNT. If you watch the output with something like mview fragments.sdf (MarvinView from ChemAxon) it will look similar to the picture displayed.

The script can be found here: Python script for extraction of Murcko fragments

In a recent paper (How To Winnow Actives from Inactives: Introducing Molecular Orthogonal Sparse Bigrams (MOSBs) and Multiclass Winnow) I applied the Winnow algorithm to the prediction of biological activities of molecules.

I thought that it would certainly be of much use to provide the code to anybody who may want to use this methodology as well. So here we go: the source code of the inplementation can be downloaded here: multiclass winnow in C++, with MySQL support (covered by the GNU General Public Licence; for what that means either read the copy enclosed or read it online).

This program is provided as is without any guarantees that it will work or anything else. For the structure of the underlying MySQL tables, should you wish to use that feature, there is no documentation yet. You can easily find the required information by reading through the source code in the file Winnow.h. If I get round to doing it, I will provide some more documentation in a while here on this page. If you have any questions, please feel free to contact me and I will try to help you out.

The program is capable of either reading the training and test data from text files, or otherwise from MySQL tables. It also includes the possibilities of bagging multiple classifiers, to use a thick threshold, as well as orthogonal sparse bigrams or an exhaustive enumeration of all features provided. Invoking the program without any arguments will display a list of all available options.

There are still a couple of very useful things that can be done with this algorithm in the area of cheminformatics. I won’t spend an excessive amount of time pushing that project any further, therefore I would be happy to share some ideas with people who may want to collaborate on a project.

I would be more than happy to receive any comments or suggestions. In particular, if there is anybody out there who would be willing to write a graphical interface, then I would be very happy to help out as much as I can. But I do not have the time to focus on that for the time being.