MultiSubstructure Matcher

richards99 · May 8, 2011, 12:16pm

Hi,

Is it possible to have a Substructure Matcher nodeto accept multiple query molecules into the second port, and the node runs a substructure search on the dataset with each query molecule. You then have the option to have:

1. Union Dataset of the results, which is the total matches returned across all the query molecules (but where there are multiple molecule matches across the queries, those molecules are only in the output once, not multiple times).

2. Exclusive Dataset of the results, which is the total matches returned across all the query molecules, minus any molecules which matched in ALL the query molecules.

3. Intersection Dataset of the results, which is the molecules which are matched in ALL the query molecule searches only.

Thanks,

Simon.

dpavlov · May 9, 2011, 7:03pm

Simon,

There are some problems in defining the concept of the MultiSubstructure Matcher node well.

First, what is the Exclusive dataset if you have more than two queries? For instance, if the queries are named A, B, and C, should the result contain structures that match A and B but not C? Or it should contain only structures that match "A, not B, not C", "B, not A, not C", "C, not A, not B"?

Second, do you want to allow the matches to overlap (i.e. to be independent from each other)? If yes, then you can do everything you need by combining a few Substructure Matcher nodes, or looping over one node.

In case you need non-overlapping matches, that raises another question -- do you need minimum amount of non-overlapping matches, or maximum? Say the structure is "NCCO" and the queries are "NC", "CC", and "CO". You can have here a single match ("CC") or two matches ("NC" + "CO"). What to choose? In any case, it would be quite a problem to implement a good algorithm that would give a determined result for this task on large amount of input queries.

Best regards,

Dmitry

richards99 · May 9, 2011, 7:51pm

HI,

Thanks for the detailed response, in terms of the exclusive dataset, I was meaning the Union minus the Intersection if that makes sense. So all the hits, except the ones which came up in everyone one of the searches.

In terms of the searches, I was thinking of each being run independently and then they are combined in the background according to whether Union, Intersection, or Exclusive is wanted.

I appreciate you can do the searches with separate substructure Matcher nodes and then join together, or as you say, by doing a Loop. However, it means it becomes tricky to get the desired outcome, as the list will contain some molecules twice or more where it came up in more than one query. To get the Intersection or Union will require you to convert everything to Canonicalised Smiles and using the Reference Row Filter node to get the desired outcome of subtracting one set of molecules from another, and then converting back to Indigo to do further analysis, it all becomes a bit messy.

I hope you can see where I am coming from.

Thanks,

Simon.

mikhail.rybalkin · October 28, 2011, 3:37pm

Hello Simon,

The latest version of Substructure Matcher node works with multiple number of queries. There is an option to match all queries, or some fixed number of queries. Also there is an option to append a column with the number of queries matched.

For Union Dataset behavior you need to use default option: "Match at least 1 query".

For Intersection Dataset behavior you need to use "Match all queries" option

For Exclusive Dataset behavior you can use "Match at least 1 query", and then filter the result by the number of queries matched: greater or equal number of queires - 1.

Exclusive Dataset is not very intuitive, and I decided not to add such option. If you can propose some intuitive user interface description for all of these cases, then we might add such option explicitly.

Best regards,
Mikhail

richards99 · October 31, 2011, 11:03am

Hi Mikhail,

I'm not sure whats happened, but there now seems to be a bug in the Substructure Matcher node.

If I take a dataset of 300 molecules, and use 1 query molecule to do a substructure with, the node freezes after 6% completion.

If I process the same molecules, but process 15 at a time using "Chunk Loop Start and End" nodes, the Substructure node works, if I process 20 at a time the Substructure node freezes.

When the node freezes it also causes KNIME to freeze up resulting in the onscreen graphics going strange.

I am unsure what the problem is here, this didnt use to happen.

Simon.

richards99 · October 31, 2011, 11:07am

I'm also finding the Feature Remover node, is randomly deleting structures too leaving a Missing cell.

Something seems to be wrong with the Indigo build!!

I've uninstalled the Indigo nodes and reinstalled them but the problem remains.

Simon.

richards99 · October 31, 2011, 11:09am

These are the type of errors I am getting;