Please wait while the transcript is being prepared...
0:00
Hello and welcome to this talk.
My name is Mark Rijnbeek.
I work at the Cheminformatics
Research Group
at the European Bioinformatics
Institute in Hinxton.
That's near
Cambridge, in the UK.
My background, I'm a
database and Java developer.
I've contributed to a number
of cheminformatics projects,
such as JChemPaint,
to the molecular editor OrChem,
which is a database
search engine for Oracle.
This talk will be about
substructure searching
chemistry databases,
about similarity calculations
and similarity searching,
and about how to
create fingerprints
and how you can use them
to search databases.
0:40
Here, you can see the
content of the talk.
There are four main parts.
As an introduction, I will
introduce the various concepts
that will be detailed later on,
introduce substructure
searching briefly,
tell you what fingerprinting is,
and point out what similarity
searching entails.
Then, for substructure
searching,
we'll be looking a
bit more on that.
And then, thirdly,
how fingerprinting can help
substructure searching,
and from that we'll jump
to similarity searching,
which also can make
use of fingerprints.
1:14
First, we have to
establish what it is
we're going to talk
about, really.
Let's assume we have a dataset
of chemical structures,
molecules or compounds in
some sort of a dataset.
Realistically, it
would be a database,
but you could also store
it in file format.
It doesn't really matter.
There will be a
set of molecules,
and we want to search
those compounds.
We can do a substructure search.
The substructure search will
yield all the molecules
that contain the specified
query structure.
User draws or formalizes a
query in whatever format,
and then the cheminformatic
software will translate that
into something it
can search with.
It will scan the database
and try to find any
sort of compound
that has that particular
structure as a substructure,
so it will look for
the superstructures.
A similarity search,
on the other hand,
gives molecules
similar to the query,
ranked by their similarity.
Similarity is a bit of a
fuzzy concept, as we'll see.
Substructure searching
is more formal.
It's either a
substructure or it isn't,
whereas whether or
not it's similar
is a bit of a fuzzy topic.
You have to give it a score.
You can say, well,
it's very similar.
Let's say it's 80% similar,
or it's very dissimilar.
Or it's only 20% similar,
but we have to come to terms
with what those
percentages really mean.
