Modelling dynamic of cellular signaling using a probabilistic approach
CREATED: 200809290200 Speaker: David Hsu ** Introduction
- static: elements and interactions
- dynamics: bistability, ultrasentivity, oscillations
- traditional approach: ODE
- ODE’s are not sufficient ** no explicit representaiton of pathway structure ** false sense of high accuracy ** How to create models automatically
- structure modeling
- parameter modeling ** local descent ** simulated annealing ** evolutionary algorithms ** Modelling with Factor Graphs
- variable node corresponds to parameter
- factor node corresponds to ODE
- variable node has an associated probability distribution
- factor node has an associated joint probability distribution ** simulate the underlying ODE ** fit results and determine error measure ** convert error to probability
- encode probability distribution over the parameter values
- exploits independence assumptions suggested by pathway structure ** Using the Factor Graph model
- determine MAP parameter estimate using belief propagation (e.g. max-product algorithm)
- converges to MAP if pathway does not have loops, gives good results in practice even if there are loops ** Application
- model composition ** model components of a model using factor graphs ** combine component by sitching together factor graphs
- data integration ** each data set results in a new instance of the same factor graph ** integrate data by sitching together the instances ** apply belief propagation across instances