TU Darmstadt / ULB / TUbiblio

Sparse learning of Markovian population models in random environments

Zechner, C. and Wadehn, F. and Koeppl, H. (2014):
Sparse learning of Markovian population models in random environments.
Cornell, In: IFAC 2014, The 19th World Congress of the International Federation of Automatic Control, Promoting automatic control for the benefit of humankind, Cape Town, South Africa, 24-29 August 2014, [Online-Edition: http://arxiv.org/abs/1401.4026],
[Conference or Workshop Item]

Abstract

Markovian population models are suitable abstractions to describe well-mixed interacting particle systems in situation where stochastic fluctuations are significant due to the involvement of low copy particles. In molecular biology, measurements on the single-cell level attest to this stochasticity and one is tempted to interpret such measurements across an isogenic cell population as different sample paths of one and the same Markov model. Over recent years evidence built up against this interpretation due to the presence of cell-to-cell variability stemming from factors other than intrinsic fluctuations. To account for this extrinsic variability, Markovian models in random environments need to be considered and a key emerging question is how to perform inference for such models. We model extrinsic variability by a random parametrization of all propensity functions. To detect which of those propensities have significant variability, we lay out a sparse learning procedure captured by a hierarchical Bayesian model whose evidence function is iteratively maximized using a variational Bayesian expectation-maximization algorithm.

Item Type: Conference or Workshop Item
Erschienen: 2014
Creators: Zechner, C. and Wadehn, F. and Koeppl, H.
Title: Sparse learning of Markovian population models in random environments
Language: English
Abstract:

Markovian population models are suitable abstractions to describe well-mixed interacting particle systems in situation where stochastic fluctuations are significant due to the involvement of low copy particles. In molecular biology, measurements on the single-cell level attest to this stochasticity and one is tempted to interpret such measurements across an isogenic cell population as different sample paths of one and the same Markov model. Over recent years evidence built up against this interpretation due to the presence of cell-to-cell variability stemming from factors other than intrinsic fluctuations. To account for this extrinsic variability, Markovian models in random environments need to be considered and a key emerging question is how to perform inference for such models. We model extrinsic variability by a random parametrization of all propensity functions. To detect which of those propensities have significant variability, we lay out a sparse learning procedure captured by a hierarchical Bayesian model whose evidence function is iteratively maximized using a variational Bayesian expectation-maximization algorithm.

Publisher: Cornell
Uncontrolled Keywords: Markov chains, Markovian models, stochastic fluctuations, molecular biology, Bayesian algorithm
Divisions: 18 Department of Electrical Engineering and Information Technology > Institute for Telecommunications > Bioinspired Communication Systems
18 Department of Electrical Engineering and Information Technology
18 Department of Electrical Engineering and Information Technology > Institute for Telecommunications
Event Title: IFAC 2014, The 19th World Congress of the International Federation of Automatic Control, Promoting automatic control for the benefit of humankind
Event Location: Cape Town, South Africa
Event Dates: 24-29 August 2014
Date Deposited: 04 Mar 2015 11:44
Official URL: http://arxiv.org/abs/1401.4026
Export:

Optionen (nur für Redakteure)

View Item View Item