1. A cancer precision medicine program
driven by multi-omic profiling, analytics
and modeling
Olivier Elemento, PhD
Director, Englander Institute for Precision
Medicine
2. Talk Overview
• The cancer precision medicine program at Weill
Cornell – what have we learnt ?
• Key challenges and how we can address them
- How do we identify what treatments are most effective
in individual patients? Patient-specific avatars
- Can we rationally identify effective combination
therapies ? Modeling complex pathways
- How do we improve our understanding of disease
biology ? Integrative and single cell biology
• The need to meld analytics and experimentation
3. Precision Medicine at Weill Cornell
Advanced
cancer
patient
Highly
personalized
treatment
recommendation
Clinical sample Profiling Interpretation Recommendation
4. Return of results requires clinical assays:
Cornell CLIA-approved whole-exome
sequencing test queries >21,000 genes
Rennert et al, 2016
5. How we report results matters – precision
medicine reports
6. Cataloguing Clinically Relevant Mutations
The Precision Medicine Knowledge Base (PMKB)
Genes
Variants
Interpretations
Tumor Types
Tissue Types https://pmkb.weill.cornell.edu/ Huang et al, 2016
7. Beltran et al, 2015; Rennert et al, 2016; Pauli et al, 2017
>1,500 cancer patients sequenced so far
13. Pauli et al, 2017
Currently actionable mutations are
not as frequent as we would like
14. Key challenges
- How do we identify what treatments are most
effective in individual patients?
- Can we rationally identify effective
combination therapies ?
- How do we improve our understanding of
disease biology and improve actionability ?
24. Can we more
rationally identify
effective
combination
therapies ?
Individual molecules effective at killing some lymphoma cells
There are tens of millions of possible
combinations of 2, 3, 4, etc drugs !!!
25. What if we could create virtual disease models
of cancer cells to test combinations in silico ?
Proliferation
Lymphomas
are addicted
to the BCR
pathway
26. Du et al, 2017
Virtual disease model recapitulates
known signaling data well
27. Virtual disease model predicts synergistic
and antagonistic drug combinations
Predictions
Experiments
28. How do we improve our
understanding of disease biology
and improve actionability ?
29. Junttila et al, 2013
Complexity of the tumor micro-environment
33. New improved approaches for
immune deconvolution from bulk RNAseq
Also – CIBERSORT (Newman et al, 2016) Davide Risso
34. The Immune Response Index integrates the
immune landscape to predict immunotherapy
responders
Machine learning
(random forest)
using clinical
outcome data
Bhinder et al, 2017;
In preparation
Independent
test set
35. Junttila et al, 2013
Tumor Microenvironment,
Single cell analysis and imaging of tumors
36. What disease really looks like at
single cell resolution
B cell lymphoma
• How do cell
population
correlate with
outcomes ?
• How do cells
communicate
and can cross-
talks be
disrupted
37. Choi et al, 2015; Durrans et al, 2015
New paracrine crosstalk between
macrophage IL6 and tumor IL6R
38. Conclusions
• Patient-specific avatars enable mini n=1
clinical trials but also iterative learning
• New technologies especially single cell
technologies allow unprecedented
understanding of the disease
• Disease is complex – requires modeling and
integrative analysis
• Experimentation/measurements and
analytics need to be closely integrated
Notes de l'éditeur
To date we have enrolled and successfully sequenced 117 patients (total enrollment is over 200). We have some of these complex cancer cases coming from other institutions as listed here.