The challenge in many application domains is to discover the hidden relationships in the data and reconstruct the complex networks that originated them. In bioinformatics, this often involves the fusion of large amounts of data from different sources in order to reconstruct the networks of interaction between proteins, gene regulation networks and metabolic networks, among others. In particular, numerous methods have been developed based on classical search algorithms to discover metabolic pathways. However, these techniques are based on the searching of linear sequences of reactions that relate only two compounds, and they suffer the problem of exponential growth of the search trees when a high number of highly connected reactions and compounds is considered.
Bio-inspired metaheuristics are a family of search and optimization techniques that allow the efficient exploration of large solution spaces. An interesting feature of these methods is the ability to intelligently explore a large number of candidate solutions concurrently, that facilitates its implementation using distributed computing. In addition, they almost naturally admit the incorporation of information from the problem domain into the different operators used by these algorithms. For this reason, this project proposes the development of new bio-inspired metaheuristics for discovering relationships networks in large biological datasets. In particular, we will address the problem of synthesizing feasible metabolic pathways that simultaneously relate a set of compounds, taking into account the possible limitations in the energy and stoichiometry of the solutions found.
PICT 2014 ANPCyT
CAID 2016 UNL