Lang Li
Department of Medical and Molecular Genetics
Indiana Institute of Personalized Medicine
Center for Computational Biology and Bioinformatics
Indiana University School of Medicine
A Tamoxifen Story: background
 The selective estrogen receptor modulator tamoxifen (TAM) was first
approved in 1977 by the FDA for the treatment of women with
metastatic breast cancer and in ensuing years for adjuvant treatment of
breast cancer.
 TAM is an established hormonal treatment for all stages of estrogen
receptor (ER)-positive breast cancer and is widely used as a chemopreventive agent in women at risk for developing the disease.
 However, there is wide inter-individual variability in the clinical
efficacy and side effects of TAM: some patients may be refractory to
TAM, and a significant proportion of patients experience side effects
that include hot flashes.
(Osborne, 1998)
A Tamoxifen Story: a pharmacokinetics idea
4-hydroxy-Tam is 30-100 times more
potent than TAM in suppressing
Estrogen-dependent cell proliferation.
(Jordan et al. 1977, 1982)
CYP3A, CYP2B6, CYP2D6 are
responsible for TAM primary
metabolism.
(Lonning et al. 1992)
There were evidences of secondary TAM
metabolites, but their functions and
metabolism pathways were not clear.
(Stearns et al., 2003)
A Tamoxifen Story: a pharmacokinetics idea
CYP3A and CYP2D6 are responsible for
TAM secondary metabolism.
(Desta et al . 2004)
Endoxifen is 10 times more potent than
4-hydroxy-TAM (Johnson et al. 2004)
A Tamoxifen Story:
pharmacogenetics and drug interaction hypotheses
1. Can CYP2D6 genetic polymorphisms predict endoxifen variation and breast
cancer patient outcome?
2. Some of the breast cancer patients also take antidepressants, and many
antidepressants are strong CYP2D6 inhibitors. Will these co-medications predict
endoxifen variation and breast cancer patient outcome?
Genetic and Drug Interaction Effect on Tamoxifen Metabolism
Endoxifen/NDM plasma ratio
0.50
0.20
0.20
0.10
0.10
0.05
0.05
0.02
0.02
r²=
-1
0
1
2
0.24
3
4
0.50
0.50
0.50
0.20
0.20
0.20
0.10
0.10
0.10
0.05
0.05
0.05
0.02
0.02
0.02
Endoxifen/NDM plasma ratio
0.50
r²=
-1
-1 0
01
12
23
0.22r²= 0.43
43
4
Proposed
CYP2D6
Gene Score
System
1 CYP2D6
Gene Score
CYP2D6
Proposed
CYP2D6
GeneScore
Score
CYP2D6
Genetic
Activity
0 -1
-1
System
0.20
0.10
0.20
sma ratio
sma ratio
Genetic /Drug- Inhibition Score
0.20
0.20
0.20
Borges et al. Journal of Clinical Pharmacology 2009
0.10
0.10
0.10
0.10
CYP2D6 Functional Genotype Predicts Patient
Survival After Tamoxifen Treatment
Goetz et al. 2005, JCO.
CYP2D6 Functional Genotypes and Co-medications (CYP2D6
inhibitor) Predicts Patient Survival After Tamoxifen Treatment
EM: CYP2D6 *1/*1,
no CYP2D6 inhibitor
IM: CYP2D6 *1/*4
no CYP2D6 inhibitor
Time to Breast Cancer Relapse
Relapse Free Survival
PM: CYP2D6 *4/*4
or CYP2D6 inhibitor
Disease Free Survival
Overall Survival
Goetz et al. 2007, BCRT.
Traditional Pharmacology Approaches in
Pharmacogenomics and Drug Interaction Studies
 in vitro studies (Human liver microsome, hepatocyte,
recombinant system, and targeted tissue cells) that
determine the drug metabolism pathways and drug
inhibition effects, and sometime drug effects.
 Clinical studies that investigate genetic effect or drug
interaction effect on drug exposure change or clinical
endpoints.
 Read the literature!!! Read the literature!!!
A Computational Biologist Approaches
 Read the literature  Literature based discovery
 Clinical Study
 Large scale clinical database data mining
 in vitro studies
 System pharmacology based discovery
Literature Based Discovery
Drug A
Drug B
Enzyme E
Existing Literatures
 Percha B., Garten Y., and Altman R. B., Discovery and explanation of
drug-drug interaction via text mining. PSB, 2011.
 Segura-Bedmar I., Martinez P., and Pablo-Sanchez C. de. Using a
shallow linguistic kernel for drug-drug interaction extraction. Journal
of Biomedical Informatics, 2011, 44, 789-804.
 Segura-Bedmar, I., Martínez, P., de Pablo-Sánchez, C. (2011). "A
linguistic rule-based approach to extract drug-drug interactions from
pharmacological documents." BMC Bioinformatics 12(suppl 2): S1
 Boyce, R., Collins, C., Horn, J., and Kale, I. (2009). "Computing with
evidence Part II: An evidential approach to predicting metabolic drugdrug interactions." J Biomed Inform 42(6): 990-1003.
in silico: DDI Prediction from PubMed Based Text Mining
 Pharmacokinetics and Drug Interaction Ontology
 Pharmacokinetics and Drug Interaction Corpus
 CYP substrates and CYP inhibitors Text Mining
 CYP enzyme based DDI prediction
PK and DDI Ontology
PK Corpus
•
Single drug in vivo PK studies:
60
•
Single drug in vivo PG studies:
60
•
in vivo drug interaction studies:
218
•
in vitro drug interaction studies:
208
XML format is also available.
Term Annotation:
Term
Sentence
Sentences with Drug Name, Dose information, Enzyme Name, PK
parameter, Units, Sample size, P-value, Mechanism, Adj word, Verb,
Action work.
Clear DDI Sentence
(CDDIS)
Vague DDI Sentence
(VDDIS)
C3 or C4
DDI
IN-VIVO DDI
IN-VITRO DDI
DDI
ADDI
DEI
ADD
I
ADEI
Non-DDI
NonDDI
NonDEI
PMID
20012601
DDI sentence
Relationship and commend
The pharmacokinetic parameters of verapamil were Because of the words,
significantly altered by the co-administration of “significantly”, (Verapamil,
lovastatin compared to the control.
20209646
lovastatin) is a DDI.
The clearance of mitoxantrone and etoposide was Because of the fold changes were
decreased by 64% and 60%, respectively, when less than 0.67, (mitoxantrone,
combined with valspodar.
valspodar.) and (etoposide,
valspodar) are DDIs.
20012601
The (AUC (0-infinity)) of norverapamil and the Because of the words, “ not
terminal half-life of verapamil did not significantly significantly changed”,
changed with lovastatin coadministration.
(verapamil , ovastatin) is a
NDDI.
13129991
The mean (SD) urinary ratio of dextromethorphan The change in PK parameter is
to its metabolite was 0.006 (0.010) at baseline and more than 1.5 fold but P-value
0.014 (0.025) after St John’s wort administration is >0.05. Thus,
(P=.26)
(dextromethorphan , St John’s
wort) is an ADDI.
19904008
The obtained results show that perazine at its Because of words, “potent
therapeutic concentrations is a potent inhibitor of inhibitor”, (perazine , CYP1A2)
human CYP1A2.
19230594
is a DEI.
After human hepatocytes were exposed to 10 Because of words, “not
microM YM758, microsomal activity and mRNA induced” and “slightly
level for CYP1A2 were not induced while those for induced”, (YM758, CYP1A2)
CYP3A4 were slightly induced.
and (YM758, CYP1A2) are
NDEIs.
Key Terms
DDI
sentences
DDI Pairs
Annotation
Categories
Frequencies
Drug
CYP
PK Parameter
Number
Mechanism
Change
Total words
CDDI
sentences
VDDI
sentences
Total sentences
DDI
ADDI
NDDI
DEI
ADEI
NDEI
Total Drug
Pairs
8633
3801
1508
3042
2732
1828
97291
1191
120
4724
1239
300
294
565
95
181
12399
Krippendorff's
alpha
0.953
0.921
0.905
Drug Interaction Information Extraction
Abstract
Identification
in vitro DDI
Abstracts
Clinical DDI
Abstracts
Sentence
Identification
DDI
Extraction
DDI
Relevant
Sentences
DDIs, ADDIs, NDDIs
DEIs, ADEIs, NDEIs
DDI Extraction
DDI Extraction Performance
Datasets
Precision
Recall
F-measure
in vivo DDI Training
0.67
0.78
0.72
in vivo DDI Testing
0.67
0.79
0.73
in vitro DDI Training
0.51
0.59
0.55
in vitro DDI Testing
0.47
0.58
0.52
Predict Potentially Interacting Drug Pairs Using Text
Mining of PubMed Abstracts
Text mining of PubMed abstracts for in vitro drug metabolism studies
involving major CYP450 isoforms using FDA recommended probes
and HLMs or recombinant CYP450s
1A2
2A6
2B6
2C8
2C9
2C19
FDA recommended
CYP2C9 inhibitors
2E1
…
…
+
2D6
Substrates
Inhibitors
3A
Predicted
potentially+
interacting
drug pairs
FDA recommended
CYP2C9 substrates
13,197 Predicted Potentially Interacting Drug Pairs
Large Scale Database DDI Data Mining
 Tatonetti, NP, Denny, J.C., Murphy, S.N., Fernald, G.H., Krishnan, G., Castro,
V., Yue, P., Tsau, P.S., Kohane, I., Roden, D.M., Altman, R.B. (2011).
"Detecting drug interactions from adverse-event reports: interaction
between paroxetine and pravastatin increases blood glucose levels."
Journal of Clinical Pharmacology and Therapeutics 90(1): 133-142.
 Tatonetti, N. P., Fernald, G.H., Altman, R.B. (2012). "A novel signal detection
algorithm for identifying hidden drug-drug interactions in adverse event
reports." J Am Med Inform Assoc 19(1): 79-85.
 Tatonetti NP, Ye PP, Daneshjou R, Altman RB. 1. Data-driven prediction of
drug effects and interactions. Science Translational Medicine, 2012 Mar 14,
4 (125).
Observational Medical Outcome Partnership - Common Data Model
 Data
 Medication Data
 Diagnosis
 Lab Tests
 2.2 Million De-identified Patient Data in the Indiana Patient Care Network
(2004 – 2009)
Drug Safety Outcome – Myopathy CDM Code (54 items)
Myositis
Muscle weakness
Polymyositis
Myoglobinuria
Myositis unspecified
[D]Myoglobinuria
March myoglobinuria
Idiopathic myoglobinuria
Exertional rhabdomyolysis
Rhabdomyolysis
Traumatic rhabdomyolysis
Non-traumatic
rhabdomyolysis
Rhabdomyolysis
Myopathy, unspecified
Myopathy, unspecified
Myalgia and myositis,
unspecified
Muscle weakness (generalized)
Polymyositis
Myoglobinuria
Rhabdomyolysis
Other myopathies
Toxic myopathy
Antilipemic and antiarteriosclerotic
drugs causing adverse effects in
therapeutic use
Myoglobinuria
Rhabdomyolysis
Polymyositis
Muscle Weakness
Myositis
Muscle Weakness
Myoglobinuria
Myoglobinuria
Polymyositis
Polymyositis
Myopathy toxic
Myopathy toxic
Muscle weakness conditions
Myositis
Myositis-like syndrome
Myopathy
Rhabdomyolysis
Myositis
Myositis-like syndrome
Muscle weakness
Generalised muscle weakness
Generalized muscle weakness
Myopathy
Myopathy, unspecified
Rhabdomyolysis
Rhabdomyolysis-induced renal failure
Myalgia and myositis, unspecified
Antilipemic and antiarteriosclerotic drugs
causing adverse effects in therapeutic use
Myopathy unspecified
Mylagia and myositis unspecified
Muscle weakness
Pharmaco-epidemiologic Study Design
Baseline co-Meds
(confounder)
Baseline Exposure
Window
D1 + D2
D1 only
D2 only
No D1/D2
Myopathy
Intermediate co-Meds
(Modifier)
Drug
Exposure
Window
Causal Inference
 Propensity Score (Donald B. Rubin, 1981)
 Propensity score construction: multinomial logistic
regression.
 Case control selection based on matched propensity
score.
 Control group selection sensitivity analysis.
 Inverse Weighted Method (James M. Robins, 1999)
 Propensity score construction: multinomial logistic
regression.
 Inverse weighted based regression
Identify DDIs Associated with Increased Risk of
Myopathy Using Electronic Medical Records
Myopathy
Risk
Synergistic Effect Model
Risk 1
0.03
+
Risk 2
0.05
<
Risk 12 ?
0.13
• Logistic regression
• Adjust for age, sex, and
medication frequency
• Drugs that treated pain were
removed.
Loratadine Simvastatin Loratadine
Only
Only
+
Simvastatin
• Bonferroni corrected
Predicted CYP450 Pathways Based DDIs
and Their Associations with Myopathy Risk
(p-value < 0.01)
Six Significant DDI Pairs
drug 1
drug 2
myopathy myopathy combined Relative
risk 1
risk 2
risk
Risk
P-value
loratadine
alprazolam
0.07
0.03
0.16
1.56
1.06E-09
loratadine
duloxetine
0.14
0.03
0.28
1.56
7.43E-09
loratadine
omeprazole 0.03
0.06
0.13
1.33
4.45E-07
loratadine
simvastatin
0.03
0.05
0.13
1.60
4.75E-07
promethazine
tegaserod
0.03
0.07
0.21
2.20
1.28E-05
loratadine
ropinirole
0.03
0.12
0.31
2.05
1.27E-05
Removed drug pairs involving drugs used to treat pain, including: chloroquine,
hydroxychloroquine, acetaminophen, oxycodone, hydrocodone, fentanyl,
tizanidine
Duke J., Han X., Wang Z., et al, 2012 PLoS Computational Biology
Do the identified drugs inhibit CYP enzymes?
CYP Enzyme
Substrate
Inhibitor
Inhibitor concentration high to low
1
2
3
4
5
6
7
control
8
A
B
Test inhibitor C
1
D
Test inhibitor E
2
F
Test inhibitor G
3
H
Positive
inhibitor

Estimate IC50 by fitting to
9
Fluorescent
metabolite
+
+
+
Cofactors
10
Backgro
und
11
12
37C for 15~45 min
Summary of Metabolic and Inhibitory Profiles
Metabolic pathway
1A2
2B6
2C9
2C19
2D6
IC50
3A
1A2
2B6
2C9
2C19
2D6
3A4
Duloxetine
Loratadine
Ropinirole
Promethazine
Simvastatin
Tegaserod
Not/Unknown
Minor
Major
IC50 > 100 uM or ND
20uM < IC50 < 100 uM
IC50 < 20 uM
Metabolism Based Inhibition Interpretation
of Six DDI Pairs
Drug 1
Drug 2
pathways
metabolism
inhibition
DDI Prediction
loratadine
alprazolam
CYP3A4
major
moderate
Moderate
loratadine
duloxetine
CYP2D6
minor
strong
Moderate
loratadine
simvastatin
CYP3A4
major
strong
Strong
promethazine
tegaserod
CYP2D6
major
strong
Strong
loratadine
ropinirole
CYP3A
minor
strong
moderate
System Pharmacology: A trans-eQTL Analysis in identify a-SNPs for CYP2D6
Gene
Expression
Genetic
Variation
CYP2D6
Enzyme
Activity
SNP (Affymetrix)
204
Gene
Expression
466
180
Enzyme
Activity
488
Samples Publically Available
SNP (Illumina)
207
Gene
Expression
466
167
Enzyme
Activity
488
Samples Publically Available
Mediation Analysis
A mediation analysis method was developed to assess the
indirect SNP effects to CYP2D6 activity mediated by gene
expressions. The mediated effect is estimated by product of
coefficients
Functional categories of Mediators
Type
cytokine
growth factor
ligand-dependent nuclear receptor
translation regulator
transmembrane receptor
ion channel
phosphatase
G-protein coupled receptor
peptidase
kinase
transporter
transcription regulator
enzyme
other
#genes
(Affy)
5
5
6
6
11
13
15
17
31
52
76
80
245
382
#genes
(illum)
7
11
8
10
14
18
22
20
39
63
127
113
365
609
What have we learned?
 The new translational biomedical information research paradigm
works!
 Literature Based DDI Discovery
 EMR data based validation
 in vitro validation
 System pharmacology based discovery
What is a drug interaction?
Drug Interaction Evidences
DDI changes
Efficacy and
ADE!
DDI changes
Drug ADME
in vivo!
DDI changes
Drug ADME
in vitro!
Drug Interaction Mechanisms
It is a
drug target
based DDI!
It is a drug
transporter
based DDI!
It is a drug
metabolism
based DDI!
Why do we care about all the information?
 Only knowing the clinical effect of a DDI won’t help prevent the DDI.
For example, polypharmacy.
 Only knowing the mechanisms of a DDI won’t be enough to
understand its clinical impact.
 Even we understand both the mechanism and clinical effect of a DDI,
we will have to worry about implementation.
We all need each other!
Wanqing Liu
Medicinal Chemistry
Purdue
David A. Flockhart
Clinical Pharm, IUSM
Metabolism
CYP CYP
Jeffrey S. Elmendorf
Physiology, IUSM
Myocyte
CYP
Liver
Portal vein
CYP
CYP
Sara K. Quinney
OBGYN, IUSM
Urine
Richard B. KimCYP
Pharmacology
UNIVERSITY of WESTERN ONTARIO
Feces
We all need each other!
PK Ontology
Drug Interaction
Text Mining
Myopathy
Definition
In CDM
Drug Interaction
Pharmacoepidemiology
Study Design
Lang Li
Luis M. Rocha
Jon Duke
Xiaochun Li
The true heroes!
PK Knowledge Database
DDI corpus
Abhinita Subhadarshini
M.S. Bioinformatics
DDI Corpus
DDI text mining
Shreyas Karnik
M.S. Bioinformatics
Pharmacogenetics
Corpus
Santosh Philips
Ph.D. Bioinformatics
Transport Ontology
Chienwei Chiang
Ph.D. Bioinformatics
EMR data processing
PK data text mining
Zhiping Wang
Ph.D. Computer Science
Ontology Construction
in vitro validation
Hengyi Wu
Ph.D. Bioinformatics
Xu Han
Ph.D. Pharm and Tox
DDI Graphical Presentation
Hrishikesh Lokhande
M.S. Bioinformatics
Funding support are from
NIGMS, AHRQ, and
IUCRG.
Thank you!