Transcript Computational biology in drug discovery

COMPUTATIONAL BIOLOGY IN DRUG DISCOVERY
RAM SAMUDRALA
ASSOCIATE PROFESSOR
UNIVERSITY OF WASHINGTON
How can we computationally screen compounds against protein
structure targets to discover inhibitors with high affinity?
MOTIVATION
Drug discovery as undertaken by the pharmaceutical company is time
consuming and expensive, with very low hit rates for the amount of
resources expended.
Computational screening of compounds against structures of protein targets
offers a way to speed up discovery time and reduce costs, but such
techniques have typically had low accuracy and need high resolution
structures.
We will capitalise on advances in computational protein structure prediction
and protein docking to improve accuracy of target-based in silico compound
screening.
Methods for obtaining structure
Experimental
Computational
X-ray crystallography
NMR spectroscopy
De novo prediction
Homology modelling
Critical Assessment of Structure Prediction (CASP)
Pre-CASP
CASP
Bias towards known structures
Blind prediction
CASP6 prediction (model1) for T0281
4.3 Å Cα RMSD for all 70 residues
http://protinfo.compbio.washington.edu/protinfo_abcmfr
Ling-Hong Hung/Shing-Chung Ngan
CASP6 prediction (model1) for T0271
2.4 Å Cα RMSD for all 142 residues (46% identity)
http://protinfo.compbio.washington.edu/protinfo_abcmfr
Tianyun Liu
Prediction of HIV protease-inhibitor binding energies with dynamics
Can predict resistance/susceptibility to six FDA approved inhibitors
with 95% accuracy in conjunction with knowledge-based methods
http://protinfo.compbio.washington.edu/pirspred/
Ekachai Jenwitheesuk
Drug discovery – current approach
Pink et al, September 2005
Drug discovery – our approach
Pink et al, September 2005
Computational protein docking with molecular dynamics protocol enables in
silico discovery of compounds that inhibit multiple targets and diseases.
Screen library of FDA approved or experimental
compounds using docking with dynamics protocol
Disease A
Disease B
Disease C
Disease X
Protein A…
Protein A2
Protein A1
1…
2…
3…
4 Inhibitor A
5…
6…
Protein B…
Protein B2
Protein B1
1…
2 Inhibitor B
3…
4…
5…
6…
Protein C…
Protein C2
Protein C1
1…
2…
3…
4…
5 Inhibitor C
6…
Protein X…
Protein X2
Protein X1
1…
2…
3 Inhibitor X
4…
5…
6…
...........
Binding affinity calculation using docking with dynamics protocol
Multi-target multi-disease therapeutic discovery
Disease A
•Protein A1
•Protein A2
•Protein A3
•…
•…
Disease B
•Protein B1
•Protein B2
•Protein B3
•…
•…
Disease C
•Protein C1
•Protein C2
•Protein C3
•…
•…
Disease X
•Protein X1
•Protein X2
•Protein X3
•…
•…
Ekachai Jenwitheesuk
Multi-target inhibition of herpesvirus proteases
Michael Lagunoff
Multi-target inhibition of herpesvirus proteases
HSV
CMV
KHSV
Ekachai Jenwitheesuk
Multi-target inhibition of herpesvirus proteases
Our best prediction showed inhibitory activity against all three classes of
herpesviruses (alpha, beta, and gamma) in cell culture, and is the only
inhibitor known to do so. We have repeated experiments several times.
Inhibition of viral growth is comparable to or better than known anti-herpes
drugs in the market (acyclovir, gancylovir, foscarnet).
Growing HSV in the presence of acylovir for a few days and measuring virus
titer results in almost no reduction with and without drug, indicating growth of
drug resistant virus. Our inhibitor continues to work well under the same
conditions.
Using low (sub-optimal) doses of both acyclvir and our inhibitor together
results in much better inhibition than either alone. Higher doses result in the
best inhibition we have observed.
Circumstantial evidence that our inhibitor does work against protease.
Inhibitory constant measurements and mouse studies are underway.
Multi-target inhibition of herpesvirus proteases
All these three viruses cause life-threatening diseases in
immunocompromised patients.
HSV drugs alone represent a > $2 billion dollar yearly market and growing at
a 10% rate. Nearly 90 million people worldwide are infected with the genital
herpes virus, and about 25 million of them suffer frequent outbreaks of
painful blisters and sores.
CMV is a major cause of mortality in transplant patients, and drugs against it
represent a $300 million dollar yearly market.
Acylovir and related drugs are all nucleoside analogues/inhibitors whose
patents will soon expire. Our protease inhibitor is a novel type of anti-herpes
agent that may be used in combination therapy.
The inhibitor has been evaluated in mouse models of cancer and found to
very nontoxic.
Topical applications are therefore possible with a high likelihood of success.
Multi-target inhibition of Plasmodium falciparum proteins
Ekachai Jenwitheesuk/Wesley Van Voorhis
Multi-target inhibition of Plasmodium falciparum proteins
We experimentally evaluated 16 of our top predictions against P. falciparum
in cell culture. 6/16 had an ED50 of  1 M, with the best inhibitor having an
ED50 of 127nM.
A negative control of 5 randomly selected compounds predicted to not inhibit
our fourteen targets did not inhibit P. falciparum growth.
Chong et al.1 experimentally screened 2687 compounds and found 87
inhibitors against P. falciparum. Weisman et al.2 screened 2162 compounds
found 72 inhibitors. Their hit rates are 3.2% (87/2687) and 3.3% (72/2162).
We are thus able to obtain a much higher hit rate of 38% (6/16) for a fraction
of the cost: Only 16 compounds costing ~$1000 needed to be tested.
Computation is fully automated and takes only a few days.
Examining overlap between our computational library and their experimental
libraries resulted in 75 compounds of which we would have tested 15. 8/15
inhibitors had an ED50 of  1M, resulting in a hit rate of 53%.
1Nat
Chem Biol 2: 415-6, 2006.
Biol Drug Des 409-16, 2006.
2Chem
Ekachai Jenwitheesuk/Wesley Van Voorhis
Other work and future directions
Our predicted inhibitors against the dengue virus are more efficacious in cell
culture than previously identified inhibitors
We have predicted inhibitors against more than 100 protein targets for over
20 diseases, including HIV, SARS, Leishmania, Tuberculosis, and Influenza.
Experimental testing is underway against some of the pathogens
responsible.
Computationally screen structurally-related compounds to experimentally
verified inhibitors from a much larger library of 1 million compounds.
Use data from experimental studies to figure out when our predicted
inhibitors are likely to be cell-active and drug-like in their behaviour; use
machine learning approaches to learn from compound characteristics (PK,
ADME, toxicity), importance of protein targets, predicted binding energies
and experimental inhibition.
Works due to the use of a combination of knowledge- and biophysics-based
methods for computational simulation.
Acknowledgements
Current group members:
Collaborators:
•Baishali Chanda
•Brady Bernard
•Chuck Mader
•David Nickle
•Ersin Emre Oren
•Ekachai Jenwitheesuk
•Gong Cheng
•Imran Rashid
•Jason McDermott
•Jeremy Horst
•Ling-Hong Hung
•Michal Guerquin
•Rob Brasier
•Rosalia Tungaraza
•Shing-Chung Ngan
•Siriphan Manocheewa
•Somsak Phattarasukol
•Stewart Moughon
•Tianyun Liu
•Weerayuth Kittichotirat
•Zach Frazier
•Kristina Montgomery, Program Manager
•Michael Lagunoff
•Wesley Van Voorhis
•Roger Bumgarner
Funding agencies:
•National Institutes of Health
•National Science Foundation
•Searle Scholars Program
•Puget Sound Partners in Global Health
•UW Advanced Technology Initiative
•Washington Research Foundation
•UW TGIF
Advantages of our approach
Costs are reduced:
Computational discovery
Use of preapproved drugs
Lower number of failed drugs
Probabily of success is higher:
Multi-target inhibition
Mechanism of action is understood
Use of preapproved drugs
Side effects may be predicted