Limits...
Efficient inference of parsimonious phenomenological models of cellular dynamics using S-systems and alternating regression.

Daniels BC, Nemenman I - PLoS ONE (2015)

Bottom Line: Simple linear models are computationally efficient, but cannot incorporate these important nonlinearities.The approach is tested by inferring the dynamics of yeast glycolysis from simulated data.With little computing time, it produces dynamical models with high predictive power and with structural complexity adapted to the difficulty of the inference problem.

View Article: PubMed Central - PubMed

Affiliation: Center for Complexity and Collective Computation, Wisconsin Institute for Discovery, University of Wisconsin, Madison, WI 53715, USA.

ABSTRACT
The nonlinearity of dynamics in systems biology makes it hard to infer them from experimental data. Simple linear models are computationally efficient, but cannot incorporate these important nonlinearities. An adaptive method based on the S-system formalism, which is a sensible representation of nonlinear mass-action kinetics typically found in cellular dynamics, maintains the efficiency of linear regression. We combine this approach with adaptive model selection to obtain efficient and parsimonious representations of cellular dynamics. The approach is tested by inferring the dynamics of yeast glycolysis from simulated data. With little computing time, it produces dynamical models with high predictive power and with structural complexity adapted to the difficulty of the inference problem.

No MeSH data available.


Related in: MedlinePlus

Testing different model hierarchies using the yeast oscillator model.As in Fig. 2, here we plot the mean correlations of predicted and true derivative values on out-of-sample test data versus the amount of data, N. We compare different orderings of adding parameters in the model hierarchy. The variance in performance over different orderings of parameters in a random hierarchy (standard deviations plotted in green) is generally smaller than the variance caused by different random realizations of the training data for the single nearest neighbor ordering used in Fig. 2 (standard deviations plotted in orange).
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4373916&req=5

pone.0119821.g004: Testing different model hierarchies using the yeast oscillator model.As in Fig. 2, here we plot the mean correlations of predicted and true derivative values on out-of-sample test data versus the amount of data, N. We compare different orderings of adding parameters in the model hierarchy. The variance in performance over different orderings of parameters in a random hierarchy (standard deviations plotted in green) is generally smaller than the variance caused by different random realizations of the training data for the single nearest neighbor ordering used in Fig. 2 (standard deviations plotted in orange).

Mentions: Finally, we test the effects of changing the choice of the model hierarchy, corresponding to changing the order of adding parameters to the model. Fig. 4 demonstrates that the variance due to the choice of the hierarchy is generally smaller than the variance due to the randomness in the sampling of the training data. Thus while restricting the model search to some single hierarchy is important, the specific choice of hierarchy is not crucial.


Efficient inference of parsimonious phenomenological models of cellular dynamics using S-systems and alternating regression.

Daniels BC, Nemenman I - PLoS ONE (2015)

Testing different model hierarchies using the yeast oscillator model.As in Fig. 2, here we plot the mean correlations of predicted and true derivative values on out-of-sample test data versus the amount of data, N. We compare different orderings of adding parameters in the model hierarchy. The variance in performance over different orderings of parameters in a random hierarchy (standard deviations plotted in green) is generally smaller than the variance caused by different random realizations of the training data for the single nearest neighbor ordering used in Fig. 2 (standard deviations plotted in orange).
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4373916&req=5

pone.0119821.g004: Testing different model hierarchies using the yeast oscillator model.As in Fig. 2, here we plot the mean correlations of predicted and true derivative values on out-of-sample test data versus the amount of data, N. We compare different orderings of adding parameters in the model hierarchy. The variance in performance over different orderings of parameters in a random hierarchy (standard deviations plotted in green) is generally smaller than the variance caused by different random realizations of the training data for the single nearest neighbor ordering used in Fig. 2 (standard deviations plotted in orange).
Mentions: Finally, we test the effects of changing the choice of the model hierarchy, corresponding to changing the order of adding parameters to the model. Fig. 4 demonstrates that the variance due to the choice of the hierarchy is generally smaller than the variance due to the randomness in the sampling of the training data. Thus while restricting the model search to some single hierarchy is important, the specific choice of hierarchy is not crucial.

Bottom Line: Simple linear models are computationally efficient, but cannot incorporate these important nonlinearities.The approach is tested by inferring the dynamics of yeast glycolysis from simulated data.With little computing time, it produces dynamical models with high predictive power and with structural complexity adapted to the difficulty of the inference problem.

View Article: PubMed Central - PubMed

Affiliation: Center for Complexity and Collective Computation, Wisconsin Institute for Discovery, University of Wisconsin, Madison, WI 53715, USA.

ABSTRACT
The nonlinearity of dynamics in systems biology makes it hard to infer them from experimental data. Simple linear models are computationally efficient, but cannot incorporate these important nonlinearities. An adaptive method based on the S-system formalism, which is a sensible representation of nonlinear mass-action kinetics typically found in cellular dynamics, maintains the efficiency of linear regression. We combine this approach with adaptive model selection to obtain efficient and parsimonious representations of cellular dynamics. The approach is tested by inferring the dynamics of yeast glycolysis from simulated data. With little computing time, it produces dynamical models with high predictive power and with structural complexity adapted to the difficulty of the inference problem.

No MeSH data available.


Related in: MedlinePlus