Jonas BÉAL
PhD Thesis
Summary
Resumé
Remerciements
Preface
I Cells and their models
1
Scientific modeling: abstract the complexity
1.1
What is a model?
1.1.1
In your own words
1.1.2
Physical world and world of ideas
1.1.3
Preview about cancer models
1.2
Statistics or mechanistic
1.2.1
The inside of the box
1.2.2
A tale of prey and predators
1.3
Simplicity is the ultimate sophistication
2
Cancer as deregulation of complex machinery
2.1
What is cancer?
2.2
Cancer from a distance: epidemiology and main figures
2.3
Basic molecular biology and cancer
2.3.1
Central dogma and core principles
2.3.2
A rogue machinery
2.4
The new era of genomics
2.4.1
From sequencing to multi-omics data
2.4.2
State-of-the art of cancer data
2.5
Data and beyond: from genetic to network disease
3
Mechanistic modeling of cancer: from complex disease to systems biology
3.1
Introducing the diversity of mechanistic models of cancer
3.2
Cell circuitry and the need for cancer systems biology
3.3
Mechanistic models of molecular signaling
3.3.1
Networks and data
3.3.2
Different formalisms for different applications
3.3.3
Some examples of complex features
3.4
From mechanistic models to clinical impact?
3.4.1
A new class of biomarkers
3.4.2
Prognostic models
3.4.3
Predictive models
3.4.4
Mechanistic models, interventions and causality
II Personalized logical models of cancer
4
Logical modeling principles and data integration
4.1
Logical modeling paradigms for qualitative description
4.1.1
Regulatory graph and logical rules
4.1.2
State transition graph and updates
4.1.3
Tools for logical modeling
4.2
The MaBoSS framework for logical modeling
4.2.1
Continuous-time Markov processes
4.2.2
Gillespie algorithm
4.2.3
A stochastic exploration of model behaviours
4.2.4
From theoretical models to data models?
4.3
Data integration and semi-quantitative logical modeling
4.3.1
Build the regulatory graph
4.3.2
Define the logical rules
4.3.3
Validate the model
5
Personalization of logical models: method and prognostic validation
5.1
From one generic model to data-specific models with PROFILE method
5.1.1
Gathering knowledge and data
5.1.2
Adapting patient profiles to a logical model
5.1.3
Personalizing logical models with patient data
5.2
An integration tool for high-dimensional data?
5.2.1
Biological relevance in cell lines
5.2.2
Validation with patient data
5.2.3
Perspectives
6
Personalized logical models to study an interpret drug response
6.1
One step further with drugs
6.1.1
Modeling response to cancer treatments
6.1.2
An application of personalized logical models
6.1.3
A pan-cancer attempt
6.2
Case study on BRAF in melanoma and colorectal cancers
6.2.1
Biological and clinical context
6.2.2
A logical model centred on BRAF
6.2.3
Cell lines data
6.2.4
Validation of personalized models using CRISPR/Cas9 and drug screening
6.2.5
Comparison of the mechanistic approach with machine learning methods
6.3
Application on prostate cancer study and challenges
6.4
Limitations and perspectives
III Statistical quantification of the clinical impact of models
7
Information flows in mechanistic models of cancer
7.1
Evaluation of models as biomarkers
7.1.1
Evaluation framework and general principles
7.1.2
Some frequent problems and recommended statistical tools
7.2
Processing of biological information
7.2.1
Information in, information out
7.2.2
Emergence of information in artificial examples
7.3
Reanalysis of mechanistic models of cancer
7.3.1
ODE model of JNK pathway by
Fey et al. (
2015
)
7.3.2
Personalized logical models: BRAF inhibition in melanoma and colorectal cancers
8
Clinical evidence generation and causal inference
8.1
Clinical trials and beyond
8.1.1
Randomized clinical trials as gold standards
8.1.2
Observational data and confounding factors
8.2
Causal inference methods to leverage data
8.2.1
Notations in potential outcomes framework
8.2.2
Identification of causal effects
9
Causal inference for precision medicine
9.1
Precision medicine in oncology
9.1.1
An illustration with patient-derived xenografts
9.1.2
Clinical trials and treatment algorithms
9.1.3
Computational models to assign cancer treatments
9.2
Emulating clinical trials to evaluate precision medicine algorithms
9.2.1
Objectives and applications
9.2.2
Target trials for precision medicine: definition of causal estimates
9.3
Causal inference methods and precision medicine
9.3.1
A treatment with multiple versions
9.3.2
Causal inference with multiple versions
9.3.3
Application to precision medicine
9.3.4
Alternative estimation methods
9.3.5
Code
9.4
Application to simulated data
9.4.1
General settings
9.4.2
Simulation results
9.5
Application to PDX
9.6
Limitations and perspectives
Discussion
Conclusion
Appendix
A
About datasets
A.1
Cell lines
A.1.1
Omics profiles
A.1.2
Drug screenings
A.1.3
CRISPR-Cas9 screening
A.2
Patient-derived xenografts
A.2.1
Overview of PDX data from
Gao et al. (
2015
)
A.2.2
Drug response metrics
A.3
Patients
A.3.1
METABRIC
A.3.2
TCGA: Breast cancer
A.3.3
TCGA: Prostate cancer
B
About logical models
B.1
Generic logical model of cancer pathways
B.2
Extended logical model of cancer pathways
B.3
Logical model of BRAF pathways in melanoma and colorectal cancer
B.4
Logical model of prostate cancer
C
About statistics
C.1
\(R^2\)
and beyond
C.1.1
Decomposition of
\(R^2\)
C.1.2
\(R^2\)
for survival data
C.2
Causal inference with multiple versions of treatment
C.2.1
Overall treatment effect with multiple versions of treatment (equation @ref(eq:overall-treatment-effect))
C.2.2
Treatment effect with predefined distributions of versions of treatment (equation @ref(eq:distrib-treatment-effect))
C.2.3
Inverse probability of treatment weighted (IPW) estimators for precision medicine
C.2.4
TMLE
D
Résumé détaillé
D.1
Modélisation et cancer
D.2
Des modèles logiques personnalisés de cancer
D.3
Quantification statistique de l'impact clinique des modèles
D.4
Conclusion
References
Published with bookdown
From the mechanistic modeling of signaling pathways in cancer to the interpretation of models and their contributions: clinical applications and statistical evaluation
Discussion
Global perspectives beyond this work