REGULATORY NETWORK MODELLING: CORRELATION FOR STRUCTURE AND PARAMETER OPTIMISATION

摘要

Due to the limitations of available gene expressionrndata, (i.e. noise and size of time series), modelling genernregulatory networks is still restricted, especially in terms ofrntheir quantitative analysis. To date, the only criterion usedrnfor model evaluation is the residual error between observedrnand simulated data. This does not assign good fitness tornmodels that can simulate the general oscillation, but arernshifted with respect to observed data. Given that oscillatoryrnbehaviour of such complex systems is mostly driven by therntopology of regulatory networks, these models may containrnimportant information on network structure, which canrnshed light on evolutionary parameter optimisation. In consequence,rna second model evaluation criterion is introducedrnhere, namely the Pearson correlation coefficient betweenrnsimulated and observed time series, which enables good fitrnto be assessed for candidate solutions able to approximaternthe general behaviour seen in the data. This is employed inrna nested optimisation algorithm, which separately analysesrnthe structure and parameters of the models. The methodrnis evaluated using both synthetic and real microarray genernexpression data, (Yeast cell cycle), and results show thatrnmodels obtained in this way display more plausible connections,rnalso contributing to simulation of quantitative behaviour.