Notes 20100330 CS 683 Structural Bioinformatics CS683 Drug Design

Jump to: navigation, search

Drug Design

Lipinksi's rule of five (recall)
little changes mean a lot in drugs-- pharmacophores
xanthine backbone (think nucleotide backbone, steroid backbone, benzene backbone, amino acid backbone etc.)
caffeine has a xanthine backbone
theme: using a scaffold, we hook on R-groups
history of drug discovery...
serendipity: chlordiazepoxide, aspartame, ether, acetylsalicylic acid, dicoumarol etc.
clinical observations: warfarin, LSD (hallucinogenic instead of cardoivascular activity), iproniazid (antidepressant instead of tuberculostatic activity) etc.
natural products: quinine, taxol, morphine, digitaline, etc.

1960~1980: structure-activity relationships: structures with the same scaffold; side chains are placed into a formula
1980~1995: rational design: ligand or protein based
1995~2010: combinatorial chemistry, genomics, proteomics; high-throughput screening

protein structure-based design...
pharmacophores are picked first -- a configuration of nearby n-atoms in space.
e.g. in a 3-point pharmacophore each point is one of the following...
a positively charged atom
a negatively charged atom
a hydrogen-bond donour
a hydrogen-bond acceptor
a hydrophobic atom
... there are 125 atom combinations in these pharmacophores; we measure the distances between each of the atoms
given a specific target, we characterize all of the pharmacophores that interact with the contact surface
one can use a 4-point pharmacophores-- a tetrahedron -- 5(125) = 625 possibilities of atoms-- there are 6 lengths in this distance geometry
similarity detection becomes difficult; use quantization -- e.g. binned by rounding at .5Å. ...?
hashed fingerprints are used often-- binary flags indicating presence or absence of a pharmacophore; a descriptor
machine learning issue: because descriptors are so long, we need at least an order of magnitude more exemplars
feature space reduction needed (a.k.a. feature selection) etc.; the curse of dimensionality
notice, this is a different way to create cheminformatics descriptors
the best presumption in the conformational analysis is actually minimal energy-- but without the protein scaffold!
similarity search (partial least squares, genetic algorithms, neural networks etc.) ...
3D databases -- look this up... ACD, CSD, NCI MDDR ...

protein-based design of protein mimic function of short protein
interacts with receptor of protein-protein interaction.
recognize that this usually takes place over a small sequence est. 3~15aa
protease inhibitors excellent target to start this approach
HIV protease example...
HIV protease is a protein that normally binds to another protein
poly-protein must be cleaved when virus matures
mimic will bind with protease and inhibit by permanently locking its sites of activity
Saquinavir is an example drug for the HIV protease
native 6aa protein has IC50 of 140nM -- with one point mutation, IC50 is 2nM, with two point mutations total, IC50 is 0.4nM
the binding affinity from a native protein was increased first 70-fold, then finally 350-fold total
other considerations ...
we don't use small peptides even though our targets accept small targets: small peptides are proteolysed
bioavailability
small peptides are rapidly metabolized: broken and cleared
establishing that a ligand is a high affinity binder is insufficient-- the above considerations must be analyzed
overarching goals...
primary: find non-peptide that has same functional capabilities
secondary (optimization): specificity, oral bioavailability, ADMET properties
achievable with non-peptides
money -- the final result will be a patentable proprietary molecule--
strategies ...
depeptidization: start with a peptide and make successive modifications (e.g. Saquinavir)
de Novo design: analysis of pharmacophores

identify the most important groups
remove atoms not contributing to functionality
add non-peptide spacer (scaffold)
Gisbert Schneider et al., Virtual screening and fast automated docking methods, Drug Design Today
two approaches...
squential growth strategy: given a protein binding pocket, place in a candidate drug fragment and incrementally add pieces of molecule until the final drug is known
fragment-placing and linking: given a protein binding pocket, start with all fragment which are needed for desired affinity and function, add linkages (scaffold).
pockets are categorized as shallow and narrow-- if there's not enough flexibility, the drug cannot get in
other strategies for filling the pocket...
mapping the pocket: hydrogen-bonded patterns-- there are directionalities associated with H-bonds (contrast: charges); deploys pharmacophores rationally
map the site by space filling with spheres: connecting the dots yields pharmacophores

the bacteria that live in the portal system (?) can also metabolize drugs into toxicants or prodrugs into active drugs (?)

canonization in organic chemistry http://www.absoluteastronomy.com/topics/IUPAC_nomenclature_of_organic_chemistry
the SAR relationship item reminds me of my own NGN modification-- scaffold-pharmacophore formula finding, similar to the one I proposed for work with Delaunay's has essentially been done back in 1960!
...but graph rewriting hasn't been done... and "pharamacophore" is a newer word
I'll have to look it up.
I'm curious to know more about pharmacophores
recall: IC50 is the concentration needed such that half a sample of target has been bound or disabled; lower means more powerful