1 / 25

Probabilistic Models that uncover the hidden Information Flow in Signalling Networks

Achim Tresch. Probabilistic Models that uncover the hidden Information Flow in Signalling Networks. Which model?. A model that explains the data merely finds associations E.g.: Epidemiology (predict colon cancer risk from SNPs). A model that explains the mechanism.

Download Presentation

Probabilistic Models that uncover the hidden Information Flow in Signalling Networks

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Achim Tresch Probabilistic Models that uncover the hidden Information Flow in Signalling Networks

  2. Which model? A model that explains the datamerely finds associations E.g.: Epidemiology (predict colon cancer risk from SNPs) A model that explains the mechanism finds explanations E.g.: Physics, Systems Biology (predict the signal flow through a cascade of transcription factors)

  3. 1st Idea Which model? Our choice: Graphical Modelsnodes correspond to physical entities, arrows correspond to interactions ? Two different types of nodes: Observable componentsPerturbed components (signals) Need for inter-ventional data

  4. How do marionettes walk?

  5. ? How do marionettes walk? This is what we observe This is the true model Both models explain the observations perfectly. What makes the right model (biologically) more plausible?

  6. Signal transmission is expensive! ? Find a consistent model with a most parsimonious effects graph 2nd Idea How do marionettes walk? This is what we observe This is the true model Both models explain the observations perfectly. Signals,Signal graph Γ Observables,Effects graph Θ What makes a model (biologically) more plausible?

  7. Nested Effects Models Signals Signal graph, Adjacency matrix Γ= (with 1´s in the diagonal) Observables Effects graph,Adjacency matrix Θ = Predicted effects Ft Parsimony Assumption: Each observable is linked to exactly one action Definition [Markowetz, Bioinformatics 2005]: A Nested Effects Model (NEM)is a model F for which F = ΓΘ

  8. Nested Effects Models Signals Why „nested“ ? If the signal graph is transitively closed, then the observed effects are nested in the sense that a→ b implies effects(a) effects(b) Observables Predicted effects Ft █ █  █ Predicted effects The present formulation of a NEM drops the transitivity requirement.

  9. Nested Effects Models Effect of signal s on observable a Signals s Ra,s a Observables Predicted effects = Ft Measured effects = Rt The final ingredient: A quantification of the measured effect strength Ra,s > 0 if the data favours an effect of s on a

  10. Nested Effects Models Assuming independent data, it follows that Note: Missing data is handeled easily: set Rs,a= 0

  11. NEM Estimation There are two ways of finding a high scoring NEM: Maximum Likelihood: Theorem (Tresch, SAGeMB 2008): For ideal data, is unique up to reversals (Corollary: if Γ is a DAG). Bayesian, posterior mode: For n≤5 signals, an exhaustive parameter space search is possible. For larger n, apply standard optimization strategies:Gradient ascent, Simulated annealingor heuristics tailored to NEMs:Module networks [Fröhlich et al., BMC Bioinformatics 2007], Triplet search [Markowetz at al., Bioinformatics 2007]

  12. Simulation R/Bioconductor package: Nessy True graphs Γ,Θ idealmeasure-ments (ΓΘ) simulatedmeasure-ments (R)

  13. Simulation True graph Estimated graph 12 edges, 212=4096 signal graphs, ~ 4seconds Distribution of the likelihoods

  14. Pathway IPathway IIsynthetic lethality Application: Synthetic Lethality • Hypotheses: • SL between two genes occurs if the genes are located in different pathways • Genes sharing the same synthetic lethality partners have an increased chance of being located in the same pathway [Ye, Bader et al., Mol.Systems Biology 2005] Pathway II Pathway I 1 a 2 b 3 • Consequence: • A gene b whose SL partners are nested into the SL partners of another gene a is likely to be located beneath a in the same pathway.

  15. Application: Synthetic Lethality Pan et al., Cell 2006

  16. Application: Synthetic Lethality 7 of 10 Genes directly linked to DNA repair Tresch, unpublished

  17. Software, References • R/Bioconductor packages: • NEM (Markowetz, Fröhlich, Beissbarth) • Nessy (Tresch) • References: • Structure Learning in Nested Effects Models. A. Tresch, F. Markowetz, to appear in SAGeMB 2008, avaliable on the ArXive • Nested Effects Models as a Means to learn Signaling Networks from Intervention Effects. H. Fröhlich, A. Tresch, F. Markowetz, M. Fellmann, R. Spang, T. Beissbarth, in preparation • Computational identification of cellular networks and pathwaysF. Markowetz, Olga G. Troyanskaya, Dennis Kostka, Rainer Spang. Molecular BioSystems, Bioinformatics 2007 • Non-transcriptional Pathway Features Reconstructed from Secondary Effects of RNA Interference.F. Markowetz, J. Bloch, R. Spang, Bioinformatics 2005

  18. Research related to the theory of NEMs • Integration of multiple data sources • Time-dependent NEMs • Allow for arbitrary signalling model • Other Research Topics • Software for data acquisition, -processing & -visualization for high-density technologies • Design and analysis of biological/clinical experiments, consulting • Teaching • Lectures & Exercises in Bioinformatics, Machine Learning, Statistics for Physicians, Group Theory, Microarray Analysis • E-learning Core Group of the Faculty • Bachelor-/Master- and PhD theses Research & Teaching Activities

  19. Acknowledgements • Florian Markowetz Lewis-Sigler Institute, Princeton • Tim Beissbarth, Holger FröhlichGerman Cancer Research Center, Heidelberg • Rainer SpangComputational Diagnostics Group, Regensburg

  20. Conclusion Exercise: Why is this administration model inefficient? Construct a model that scores better! Thank You!

  21. What I did not show … Automatic Feature Selection, without Control experiment:Estimated graph (120 genes selected)

  22. What I did not show … The „observed“ graph of the Fellmann estrogen receptor dataset

  23. What I did not show … 15 Genes 17 Knockdown Experiments 6 of them double Knockdowns

  24. What I did not show … Same Data, With prior knowledge.

More Related