Construction of large signaling pathways using an adaptive perturbation approach with...

24
signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris E. Messinis, Thomas S. Weiss, Julio-Saez Rodriguez, Leonidas G. Alexopoulos Naga Srinivas Sooraj Vedula NetID:nvedul2 Spring 2015

Transcript of Construction of large signaling pathways using an adaptive perturbation approach with...

Page 1: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Construction of large signaling pathways using an adaptive perturbationapproach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris E. Messinis, Thomas S. Weiss, Julio-Saez Rodriguez, Leonidas G. Alexopoulos

Naga Srinivas Sooraj Vedula NetID:nvedul2 Spring 2015

Page 2: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Outline

-Introduction

-Proteomic technologies

-Experimental Procedure:Day1(Collecting sample cells from patients)

-Day2(Ligand Selection and GMD)

-Day3(Combinatorial experiment, Hill function and ILP formulation)

-Results

-Research questions

Page 3: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Introduction - Cell signaling refers to how information or a message moves inside of the cytoplasm of a cell.

- Ligand(stimuli).- Signaling pathways - entire set of cell changes induced by receptor activation.- Perturbation is caused due to stimuli.

http://nikolai.lazarov.pro/lectures/2014/medicine/cell_biology/04_Cell_Hierarchy_Chemical_Composition.pdf

Page 4: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Introduction(Contd.)-Hepatocytes – liver cells that have proteins.

-Phosphoproteins – chemically bounded to phosphoric acid.

-Phosphorylation signals – signal flow initiated by key phosphoprotein.

Page 5: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Proteomic technologies 1)Technologies that make no prior assumption about the sample’s protein content.

e.g. Mass Spectrometry(MS)– breaking down to peptide level and using their sequence . Tedious.

2) Affinity based methods – response to stimuli.

e.g. xMAP technology – Using dyed spheres with different combination of different dyes. Making use of Fluorophore.

So we use xMAP technology as it can test thousands of cells and fast result generation.

Page 6: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Experiment:Day1(Patient Interaction)

- Liver tissue samples are obtained from patients with liver tumor secondary or higher degree cancer.

-Hepatocyte are isolated from samples obtained.-Primary human hepatocytes were place 96-well plates.

Source : http://www.evergreensci.com/labware-catalog/microplates-strips-and-films/uvt-acrylic-96-well-plates/http://www.luminexcorp.com/prod/groups/public/documents/lmnxcorp/reagents-beads.jpg

Page 7: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Day2:Ligand Screening and Data Acquisition

-Ligand Screening - A library of 81 stimuli was put together with specific concentrations(text mining).

e.g. cytokines, chemokines.

- 14 key phosphoproteins were chosen based upon significance of pathways involved.

- Result after exposure to laser.

Page 8: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Day2:Ligand Selection -The Gaussian Mixture Distribution (GMD) was used for ligand selection procedure . Smooth bell curve can be attributed to continuous random variable since phosphorylation activity can have multiple outcomes . Below is Probability distribution function.

Phosphorylation Activity

Frequency

AKT

Page 9: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Contd. Gaussian Mixture Model Model

Page 10: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Contd. Gaussian Mixture Model -Discretization of experimental data can be attributed to bell curve comparison in both modes.

-Discrete part- If the probability distribution function of the phosphorylation signals are compared and the one with highest frequency is state of the signal (ON or OFF).

-From Statistics Toolbox of Mat lab gmdistribution.fit() and pdf() were used.

-Ultimately 15 out of 81 stimuli that activated at least one of the signals were allowed to progress.

Page 11: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Day3-Combinatorial Experiment -Experiment is done with the same set of hepatocytes used earlier.

-Combinatorial fashion used earlier made it impossible to use the previously applied procedure.

-So hill function is used to scale the fold change of signal.

=

Page 12: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Day3-Combinatorial Experiment(contd.,)

- is the normalized measured value of species j

in experiment k

is the unstimulated measured value of species j in

experiment k,

is the stimulated measured value of species j in

experiment k,

n is the hill coefficient, herein n = 4,

p is a user defined threshold beyond which the signal is considered activated(signal to noise ratio) here its considered 2.

Page 13: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Normal Hill Function

- fraction of the ligand-binding sites on the receptor protein which are occupied by the ligand. L - free (unbound) ligand concentration  - occupied binding sites. n – Hill Coefficient(if its 1?)

Page 14: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Day3- Generic PathwayLigand/Stimuli

reactions

Phosphoproteins

Active phosphoproteins

Page 15: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Pathway pre-processing: controllability, observability and feedback loops

-Enabled using CellNetOptimizer.

-Making use of DFS we remove the feedback loops.

-Controllability and observability observed using Warshall’s algorithm and unnecessary edges are removed.

egf

egfr

shc

grb2

Page 16: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Observable-controllable pathway

Page 17: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Integer Linear Programming formulation

- Goal 1 is to find an optimal set from potential reactions superset.

- Goal 2 is to be as close to experimental results as possible. - Applied under 2 settings positive weight and negative weight.

| - Predicted value of a node (binary decision variable) (0 or 1)(constants) – measured value of node (normalized to between 0 and 1)

In each experiment a subset of species is introduced to the system and another subset is excluded from the system.(using ligands) These are summarized by the index sets and .

Page 18: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

ILP formulation - are user defined weights of nodes (for species j in experiment k)

- indicating if a Reaction is possible or not (=0 connection not present, =1 Connection present).

- are user defined weights of edges (for reaction i ).

- The first term of (1) corresponds to the measurement–prediction mismatch, and its minimization guarantees the goodness of fit of the solution. The summation is performed only over the measured species for each experiment.

- The second term of (1) if > 0 minimizes the size of the pathway, else if 0 maximizes the size of the pathway.

Page 19: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

ILP formulation Constraints:

<= where i set of reactions and k is number of experiment.

<= where j set of reactants in reaction i .

= 0 where K set of experiments and j

= 1 where K set of experiments and j

-indicate if reaction will take place (=1) or not (0) in the experiment according to the model predictions

-NP-Hard? Using constraints and tools we can reduce the time complexity to polynomial time

Page 20: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

optimized pathway conserved by ILP

Page 21: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Statistics of ILP -Earlier there were 365 reactions and we removed 204 using ILP . 53 reactions are included in minimum pathway. 161 included in maximum pathway.

Page 22: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Results -Just by using 14 phosphoprotein signals used in this study were sufficient to give a pathway coverage equal to 68.5% of the generic.

-Predicted reactions are close to experimentally observed reactions.

-Authors were able to effectively identify the cell reaction to stimuli by identifying optimal pathways.

Page 23: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

References - “Identifying Drug Effects via Pathway Alterations using an Integer Linear Programming Optimization Formulation on Phosphoproteomic Data” Alexander Mitsos, Ioannis N. Melas, Paraskeuas Siminelakis, Aikaterini D. Chairakaki, Julio Saez-Rodriguez, Leonidas G. Alexopoulos

- “Functional genomics and proteomics as a foundation for systems biology” Kunal Aggarwal and Kelvin H. Lee

- http://www.cdpcenter.org/resources/software/cellnetoptimizer /

- http://www.luminexcorp.com/TechnologiesScience/xMAPTechnology /

- “Networks Inferred from Biochemical Data Reveal Profound Differences in Toll-like Receptor and Inflammatory Signaling between Normal and Transformed Hepatocytes” Leonidas G. Alexopoulos,Julio Saez-Rodriguez,Benjamin D. Cosgrove , Douglas A. Lauffenburger and Peter K. Sorger

Page 24: Construction of large signaling pathways using an adaptive perturbation approach with phosphoproteomic data Ioannis N. Melas, Alexander Mitsos, Dimitris.

Thank You