There are a few areas of science and technology which are only as challenging, emerging and promising as computational biology. This area is looking for its mathematical foundations, for methods of prediction while guaranteeing robustness, and it is of a rigorous interdisciplinary nature. In this paper, we deepen and extend the approach of learning gene-expression patterns in the framework of gene-environment networks by optimization, especially, generalized semi-infinite optimization (GSIP). With respect to research done previously, we additionally imply the fact that there are measurement errors in the microarray technology and in the environmental data likewise; moreover, the effects which exists among the genes and environmental items can seldom be precisely quantified. Furthermore, we present the well-established matrix algebra for our extended model space, and we indicate further new approaches.