Please wait while the transcript is being prepared...
0:00
Hello, I'm Nathan Brown
from the Institute of Cancer Research
in London,
and today I'm going to be talking to you
about computational methods
in ligand-based drug design
for medicinal chemistry.
0:10
So ligand-based drug design -
it's important to understand
what we do and how we achieve it.
So we use computational methods
to design, select and prioritise
synthetic chemistry targets
that will then contribute positively
to a medicinal chemistry project.
0:23
And how we do that is to,
starting from the top,
to identify potential substituents;
if we've got a core
scaffold of interest.
We then can enumerate libraries
in the computer, virtually.
From those libraries of possible
new chemical structures,
we can calculate predictions
in the computer using
empirical models and first principles models.
Then we prioritise compounds,
and then the most time-consuming part
is the chemical synthesis
and biological testing
of those target molecules.
Followed by data analysis
of the resulting data
and then we iterate around that
to design new, more potent,
and more effective drugs.
0:57
So an overview of this talk is,
we're going to cover
molecular representations,
a number of different types
of molecular descriptors
including physicochemical properties
and structural fingerprints.
We'll then move onto concepts
of molecular similarity
and how they're used in virtual screening.
We'll then conclude with some aspects
of statistical learning,
in particular unsupervised learning
and supervised learning.
1:18
So the foundations of modern
computational chemistry
come from graph theory
that was founded as a discipline
in the early 18th century
by Leonhard Euler,
where he tried to understand
a path problem in Königsberg
where you would cross every bridge
in Königsberg only once
and pass each of the land masses.
So Euler started with a map of Königsberg
and then colour coded the landmasses,
each landmass is coloured differently,
with the bridges connecting them
highlighted in blue.
So this map already contains
a lot of redundant information
that isn't necessary
for the problem at hand.
So we can take away the actual map
and we're just left
with the morphology of landmasses
and the connectivity
between these landmasses.
Then we can abstract this even further
because we don't need to know
the morphology of the landmasses,
we just need to know that they exist,
and they're connected.
So we can get rid of morphology
and then we end up
with an abstract graph representation
which led to the foundation
of the mathematical discipline
of graph theory.
