ABSTRACT
We have used recurrent artificial neural nets (see Mjolsness et al., 1991) to simulate how genes and their interactions in cells determine the phenotypes of animal organs or simple organisms and their development, as well as how such gene interactions evolve under particular (simulated) environmental conditions or constraints: nodes in these neural nets correspond to genes and node activation levels to gene expression levels. We optimize the parameters in these models (node interaction strengths, activation and decay rates, thresholds and so on) in order to either fit experimental data (gene expression patterns) or to impart desired features to the simulated system and make it conform to constraints.
