Reconstructing biological networks, such as metabolic and signaling networks, is at the heart of systems biology. We present several equation discovery approaches to reconstructing network structure, which recover the full dynamic behavior of a network. These take as input measured time course data, as well as existing domain knowledge, such as partial knowledge of the network structure. We demonstrate the use of these approaches on tasks of rediscovering known networks and proposing models for unknown networks.
COBISS.SI-ID: 21901607
The minimal description length (MDL) principle allows us to find an optimal trade-off between the complexity of a model and its predictive error. We propose an MDL scheme for regression by polynomial equations, which includes coding schemes for polynomials and the errors they make on data. We compare this principled MDL scheme to an ad-hoc MDL scheme and show that it performs better, so that it is comparable in accuracy to other commonly used methods for regression, such as model trees, while producing much smaller models.
COBISS.SI-ID: 21912103
There is a strong interest in agriculture in gaining new knowledge about outcrossing between GM and conventional crops. Many models of outcrossing between crops have been developed, but most of them are mechanistic, very complex and rarely evaluated against real data. Our approach uses field measurements and background knowledge to develop accurate equation-based models of the outcrossing between GM and conventional maize crops. The use of background knowledge and equation-discovery is a novelty and a unique contribution to the study of outcrossing between GM and conventional maize.
COBISS.SI-ID: 22574375
The paper presents an automatically discovered model equation for predicting the concentration of the algal species dinoflagellate in Lake Kinneret. We applied an automated modelling tool (Lagramge), which integrates the knowledge- and the data-driven modelling approach. Using the data and expert knowledge coded in a modelling knowledge library, Lagramge successfully discovered several suitable mathematical models from which, based on the expert’s visual estimation and validation of the models, one optimal model capable of long-term predictions was selected.
COBISS.SI-ID: 24367399
We first apply inductive process modeling methods to induce models from measured data in a source domain (e.g., ocean aquatic ecosystem), use these models to learn constraints for inductive process modeling, and transfer the constraints to the modeling tasks in the target domain (e.g., lake aquatic ecosystem). The reported results show that that cross-domain transfer of constraints is beneficial in both directions, from ocean to lake ecosystems and vice versa. The transferred constraints increase the efficiency of learning and only slightly decrease of the predictive accuracy of the models.
COBISS.SI-ID: 3541422