Transcript Document

NOVEL PARADIGMS FOR DRUG DISCOVERY
SHOTGUN COMPUTATIONAL MULTITARGET SCREENING
RAM SAMUDRALA
ASSOCIATE PROFESSOR
UNIVERSITY OF WASHINGTON
SHOTGUN MULTITARGET DOCKING WITH DYNAMICS
ALL KNOWN DRUGS
FRAGMENT BASED
(~5,000 FROM FDA) DOCKING WITH DYNAMICS
(~50,000,000)
PRIORITISED
HITS
DISSOCIATION CONSTANTS (KD)
(~300-500)
+
ALL TARGETS WITH KNOWN
STRUCTURE (~5,000-10,000)
MACHINE
LEARNING
IN VITRO
STUDIES
herpes, malaria, dengue
hepatitis C, dental caries
HIV, HBRV, XMRV
M Lagunoff (UW), W Van Woorhis (UW),
S Michael (FCGU), J Mittler/J Mullins (UW),
G Wong/A Mason/L Tyrell (U Alberta),
W Chantratita/P Palittapongarnpim (Thailand)
CLINICAL STUDIES/APPLICATION
INITIAL CLINICAL TRIALS
IN VIVO STUDIES
PROSPECTIVE PRELIMINARY VERIFICATION
Predicted
protease (dimer) +
inhibitor:
HERPES
DENGUE
(HSV, CMV, KSHV)
Viral E protein
Observed:
Function is inactivated.
Prediction #1
Prediction #2
KD protease ligand ≤ μM
KD protease dimer ≤ μM
Herpes viral load
Experiment 1
Experiment 2
2/4 ≤ µM ED50
against dengue virus
PLoS Neglected Tropical Diseases, 2010.
14 targets
MALARIA
Multitarget protocol: 2,344 → 16 → 6 ≤ 1 µM ED50
HTS protocol:
2,687
→ 19 ≤ 1 µM ED50
HTS protocol:
2,160
→ 36 ≤ 1 µM ED50
Docking protocol: 355,000 → 100 → 1 ≤ 10 µM ED50
Docking protocol: 241,000 → 84 → 4 ≤ 10 µM ED50
Trends in Pharmacological Sciences, 2010.
SHOTGUN MULTITARGET DOCKING WITH DYNAMICS
ALL KNOWN DRUGS
FRAGMENT BASED
(~5,000 FROM FDA) DOCKING WITH DYNAMICS
(~50,000,000)
PRIORITISED
HITS
DISSOCIATION CONSTANTS (KD)
(~300-500)
+
ALL TARGETS WITH KNOWN
STRUCTURE (~5,000-10,000)
MACHINE
LEARNING
herpes, malaria, dengue
HIV, HBRV, XMRV
hepatitis C, dental caries
M Lagunoff (UW), W Van Woorhis (UW),
S Michael (FCGU), J Mittler/J Mullins (UW),
G Wong/A Mason/L Tyrell (U Alberta),
W Chantratita/P Palittapongarnpim (Thailand)
CLINICAL STUDIES/APPLICATION
Docking with dynamics
Fragment based
Multitargeting
Use of existing drugs
Drug/target maching learning matrix
PK/ADME/bioavailability/toxicity/etc.
Biophysics + knowledge iteration
Fast track to clinic (paradigm shift)
Cocktails/NCEs/optimisation
Translative: atomic → clinic
DISCOVER NOVEL OFFLABEL USES OF MAJOR THERAPEUTIC VALUE
Correlation coefficient
PROTEIN INHIBITOR DOCKING WITH DYNAMICS
HIV protease
1.0
with MD
0.5
without MD
ps
0
0.2
0.4
0.6
0.8
1.0
MD simulation time
Jenwitheesuk
ACCURACY COMPARISON
Bernard & Samudrala. Proteins (2009).
BACKGROUND AND MOTIVATION
My research on protein and proteome structure, function, and interaction is
directed to understanding how genomes specify phenotype and behaviour;
my goal is to use this information to improve human health and quality of life.
Protein functions and interactions are mediated by atomic three dimensional
structure. We are applying all our structure prediction technologies to the
area of small molecule therapeutic discovery.
The goal is to create a comprehensive in silico drug discovery pipeline to
increase the odds of initial preclinical hits and leads leading to significantly
better outcomes downstream in the clinic.
The knowledge-based drug discovery pipeline will adopt a shotgun approach
that screens all known FDA approved drug and drug-like compounds against
all known target proteins of known structure, simultaneously examining how
a small molecule therapeutic interacts with targets, antitargets, metabolic
pathways, to obtain a holistic picture of drug efficacy and side effects.
Find new uses for existing drugs that can be used in the clinic, with a focus
on third world and neglected diseases with poor or nonexisting treatments.
MULTITARGET DOCKING WITH DYNAMICS
NOVEL FRAGMENT BASED
MULTITARGET SCREENING
Disease &
target identification
TRADITIONAL SINGLE
TARGET SCREENING
COMPOUND SELECTION
Multiple disease related proteins
Compound
database (~300,000)
Single disease related protein
Compound library
DRUG-LIKE
(~5000 from FDA)
Computational docking with dynamics
Initial candidates
Experimental verification
Success rate +++++
Time .
Cost $
Computational docking
Initial candidates
Experimental verification
Success rate ++
Time
.
Cost $$$
High throughput screen
Experimental verification
Success rate +
Time
.
Cost $$$$$
INHIBITION OF ALL REPRESENTATIVE HERPES PROTEASES
Predicted:
Observed:
Function is inactivated.
protease ligand KD < μM
protease dimer KD < μM
Jenwitheesuk/Myszka
INHIBITION OF ALL HERPESVIRUSES
Viral load
HSV
KSHV
CMV
Fold inhibition
Computationally predicted broad spectrum human herpesvirus protease inhibitors is effective in vitro
against members from all three classes and is comparable or better than antiherpes drugs
Our protease inhibitor acts synergistically with acyclovir (a nucleoside analogue that inhibits replication)
and it is less likely to lead to resistant strains compared to acyclovir
HSV
Viral load
HSV
Experiment 1
Experiment 2
Experiment 3
Lagunoff
MALARIA INHIBITOR DISCOVERY
Predicted
inhibitory
constant
10-13
10-12
10-11
10-10
10-9
10-8
10-7
None
Trends in Pharmacological Sciences, 2010.
Jenwitheesuk/
Van Voorhis/Rivas/Chong/Weismann
MALARIA INHIBITOR DISCOVERY
+++++
COMPARISON OF APPROACHES
$
Multitarget computational protocol
2,344 compounds
High throughput protocol 1
2,687 compounds
simulation
16 top predictions
experiment
6 ED50 ≤ 1 μM
High throughput protocol 2
2,160 compounds
high
high
throughput
throughput
screen
screen
19 ED50 ≤ 1 μM
36 ED50 ≤ 1 μM
++
$$$$$
Computational protocol 1
241,000 compounds
simulation
84 top predictions
In comparison to other approaches, including
experimental high throughput screens, our
multitarget docking with dynamics protocol
combining theory and experiment is more
efficient and accurate.
experiment
4 ED50 ≤ 10 μM
Computational protocol 1
355,000 compounds
simulation
100 top predictions
experiment
1 ED50 ≤ 10 μM
Trends in Pharmacological Sciences, 2010.
+++
$$$
Jenwitheesuk/Van Voorhis/Rivas
DENGUE INHIBITOR DISCOVERY
Prediction #1
Prediction #2
PLoS Neglected Tropical Diseases, 2010.
Jenwitheesuk/Michael
WHY WILL IT WORK
Fragment based docking with dynamics: dynamics improves accuracy;
fragmentation exploits redundancy in existing drugs; most accurate docking
protocol out there.
Use of existing drugs: exploits all the knowledge from Pharma.
Multitargeting: multiple low Kd can work synergistically; screening for targets
and antitargets simultaneously.
Knowledge based: potential from known structures, will have a big matrix
relating drugs, targets, PK, ADME, solubility, bioavailability, toxicity, etc.; rich
dataset for combining our biophysics based methods with machine learning
tools in an iterative manner.
Known drugs
Targets with known structure
docking score, Kd, PK, ADME,
absorption, bioavailability, toxicity
BROADER IMPACT
Multiple drugs can be combined to produce therapeutic effect and overcome
disease resistance.
Good for any condition where one or more viable targets exist.
Harnesses the power of all the drug discovery done thusfar; new paradigm
for fast track FDA approval
Translational approach goes from providing atomic mechanistic detail to
measuring clinical efficacy in one shot.
Protocol can be used to design novel drugs also.
SUITABILITY FOR THE PIONEER AWARD
Not good for Pharma because of reuse of existing drugs (most profit in novel
compounds)
Not good for Pharma because of focus on third world/neglected diseases.
Not good for Pharma because of nonfocus on single target model they love.
Marked departure from my protein structure prediction work, but now applied
research from basic protein folding to producing therapeutics in a clinic.
Funding will help focus work on drug discovery which until now has been
done on a shoestring.
CONCLUSION
High risk endeavour is successful if one or more diseases
currently without an effective treatment can be treated completely.