The current bioinformatics leap is on modelling. A model is very much essential for data interpretation but, the fundamental question is: what levels of models should be chosen? A model class should be selected according to the data requirements and the objectives of the modeling
and analysis. This involves classical engineering tradeoffs. For example, a “fine” model with many parameters will capture detailed “low-level” phenomena, but will require large amounts of data for the inference, for fear of the model being “over fitted ” to the data, whereas a less complex “coarse” model with fewer parameters will capture “high-level” phenomena, but will require small amounts of data. Within a chosen model class, according to Occam’s Razor principle, the model should never be made more complex than what is necessary to “explain the data”. There are numerous approaches for modelling gene regulatory networks: it goes from linear models, Bayesian networks, neural networks, non linear ordinary differential equations, and stochastic models to Boolean models, logical networks, Petri nets, graph-based models, grammars, and process algebras.
So, which model will you choose for your biological data...?!!
and analysis. This involves classical engineering tradeoffs. For example, a “fine” model with many parameters will capture detailed “low-level” phenomena, but will require large amounts of data for the inference, for fear of the model being “over fitted ” to the data, whereas a less complex “coarse” model with fewer parameters will capture “high-level” phenomena, but will require small amounts of data. Within a chosen model class, according to Occam’s Razor principle, the model should never be made more complex than what is necessary to “explain the data”. There are numerous approaches for modelling gene regulatory networks: it goes from linear models, Bayesian networks, neural networks, non linear ordinary differential equations, and stochastic models to Boolean models, logical networks, Petri nets, graph-based models, grammars, and process algebras.
So, which model will you choose for your biological data...?!!
No comments:
Post a Comment