2. BioVariance - Overview
Two distinct offerings rooted in the same data:
• “Medical value content”-as-a-Service for Healthcare
• making sense of genomic & other data in context as
service offering
• We aim for testable
hypotheses that are well-
supported by data from
scientific databases and the
literature
Bio-Variance
BaVarians ;)
3. Background literature
Prediction of biological targets for compounds using multiple-category Bayesian models
trained on chemogenomics databases
Nidhi, Glick M, Davies JW, Jenkins JL.
J Chem Inf Model. 2006 May-Jun;46(3):1124-33.
Predicting new molecular targets for known drugs
Michael J Keiser, Vincent Setola, John J Irwin, Christian Laggner, Atheir I Abbas, Sandra J
Hufeisen, Niels H Jensen, Michael B Kuijer, Roberto C Matos, Thuy B Tran, Ryan Whaley,
Richard A Glennon, Jérôme Hert, Kelan LH Thomas, Douglas D Edwards, Brian K Shoichet,
Bryan L Roth
2009/11/1 Nature Volume 462 Issue 7270 Pages 175-181
Gaining insight into off-target mediated effects of drug candidates with a comprehensive
systems chemical biology analysis
Scheiber J, Chen B, Milik M, Sukuru SC, Bender A, Mikhailov D, Whitebread S, Hamon J,
Azzaoui K, Urban L, Glick M, Davies JW, Jenkins JL.
J Chem Inf Model. 2009 Feb;49(2):308-17.
4. At a glance
Target profile prediction
• Starting from early Drug Discovery it is very
important to understand compound activity profiles
and underlying mechanisms
• Cost restrictions render it impossible to perform a
comprehensive in-vitro testing of all compounds
against all targets
• Computational approaches help to identify the
targets having the highest probability of becoming
problems and to exclude those that will likely not
become an issue
6. Data input
• Your compounds
• Chemogenomics datasets
• Your internal data incorporated where applicable
• Specifically curated scientific papers around
particular targets (especially if some interesting
facts turn up in first run)
7. Computational description of
molecules
• Descriptor selection heavily impacts outcome of
analyses
• Depending on your main objectives different
technologies are the best fit, we will discuss this
in detail with you
0 1 0 0 2 0 0 0 1 0
8. Statistical modeling
• Activity is either modeled as yes/no or in
categories (depending on your needs)
• Plenty of positive results with naïve
Bayes, therefore method of choice
• Other technologies depending on data/on
request
• Strict model validation
9. Model Validation - Example
2
n n compounds
1
n
3 3
defined activity
training data set test data set
times
Prediction model
n
3
2
repeat 3 x
model
predict
Internal measure for model quality
R2CV-50%
10. Model Validation - Example
2 1
n n
3 3
at leaste 100 x
repeat
Prediction Model
mal
n
External measure
3
2
for model quality
repeat 3 x
R2Test,Avg
model
predict
Internal measure for model quality
R2CV-50%
11. Prediction results
T1
• Based on model sets for each target, T2
i.e. there are 100 prediction results for T3
each target
…
• These are further analyzed, usually
median predictive value taken for
prediction and ranking
…
• Result: A ranked list with associated
probabilities for each compound
12. What does the result mean?
• Targets need to be annotated with phenotypic
outcome – i.e. what does it mean that the
compound is hitting this target?
• Do we have opportunities ( repurposing) or
liabilities ( side effects) or both?
• How do different compounds compare?
• What predictions should be confirmed by testing?
13. How are targets linked to diseases? –
Data Source examples
As of 2011, 1200 human GWASs have been
published on over 400 traits
Manolio TA. N Engl J Med 2010;363:166-176.
Phenocopy effect:
If one can link a predicted target to one of these, you have a repurposing
opportunity or symptoms as possible side effects
14. Possible extensions
Diving into chemical biology
• Map into
pathways
• Retrieve
marketed drugs
and clinical
candidates that
act in these
pathways
15. Outlook
The right drug for the right patient
at the right time & right dose is only possible
if you have the right knowledge within the right context
right in place
We will further work on this!
16. Thank you for your attention!
josef.scheiber@biovariance.com
Phone: +49 – 89 – 189 6582 – 80
Garmischer Str. 4/V
80339 Munich / Germany