Inference of Linear Gene Regulatory Model –Reverse Engineering by Genetic Algorithms–

The progress on gene expression profiling technology has allowed us to venture into the underlying regulatory rules and suggest many gene network models. Among them, the quantitative models are probably the most potent to achieve the highest precision and sensitivity. But the practical limitation of the approach is the requirement of lengthy data points and vulnerability to noise in the data. With present DNA micro-array technology, the large data set is very costly, and the noise is inevitably significant in expression levels. We have focused on using a matrix influence model to represent the network regulation, in which the regulatory input to a gene's expression level is represented by the summation of the regulation state of other genes. We propose an implementation with genetic algorithms, which allows us to analyze the data sufficiently with noise and small amount of data. The method was applied to simulated and Rat Cervical Spinal Cord development data to predict the network model from the expression profile.