Notes 20100401 CS 683 Structural Bioinformatics Peptidomimetism

Jump to: navigation, search

Recall HIV Protease Inhibitors

mimic the part of the viral protein that will be cleaved by the protease-- the mimic will not be cleaved and 'glues' shut the protease
monkey wrench in a gearbox -- correct size wrench etc.
recall: Saquinavir -- clarification-- the protein fragment that Saquinavir is based on is the target of HIV protease
peptide sequences can't be used in general due to degradation etc. (see previous)...

start with x-ray crystal structures
combinatorial chemistry
methods: (1) depeptidization (replacing protein fragments until we end up with a mimic) (2) de novo design
(many slides were reviewed from last lecture, started with strategies in general ... de novo ... pocket space filling ... pharmacophore)

an example-- general algorithm...
we create a graph which describes relationships and descriptors as opposed to a literal spatial mapping
considers the pharmacophores (3-points) and also the contact surface of the binding pocket
the algorithm converges when three points in the pharmacophores overlaps three points in the protein surface
we are not looking for maximal clique; we're concerned with finding a small clique of reasonable size
we alter our 'accept' circumstances so that we run in polynomial time as opposed to NP-complete time
there are binding modes for a drug to a binding pocket -- this is a non-ideal circumstance
when drug design started, we didn't actually have an idea of modes-- the assumption is that a drug can bind exactly one away
we now know that multiple binding modes can complicate the data!
many machine learning algorithms are designed by humans with the latent assumption of a single space search -- this is philosophically incorrect, more importantly; it truly increases the dimensionality of the problem (e.g. ANN, SVM)
this is a problem that hasn't yet a solution...

Rarey, Kramer, Lengauer (1995) Time-Efficient Docking of Flexible Ligands into Active Sites of Proteins
assumes and models binding where the ligand is flexible
requires a lot of interaction with the user...
there are three phases: (1) base selection, (2) base placement, (3) construction
base selection: user selects connected part of ligand
base placement: user selects binding pocket of protein
construction: --
base selection trade-off ... need to select a relatively large region

type 1: Peptide backbone, non-standard amino acids
type 2: non-peptide molecules, mimics small peptides
type 3: non-peptide scaffold, amino-acid like side chains
non-standard aa: protein will not be degraded by proteases
amino-acid analogues... huge libraries-- goal: replace natural aa with synthetic with better ADMET properties
libraries exist for each amino acid
adding cycles stabilizes molecules -- single, double, triple ring combinations