A Bi-Level Integration Algorithm for Simultaneous Modeling of Multiple Cells

With the advent of high-throughput technologies, “omics” data types have provided quantitative data for thousands of cellular components across a variety of scales. Integration of “omics” data in a genome-scale metabolic model has been a major challenge in systems biology. In this research, a new bi-level linear optimization algorithm is introduced that limits the upper bound of the metabolic reactions using a constraint based on the gene expression. This algorithm can simultaneously integrate the expression of multiple cells into a metabolic model in a comparative manner. Outer problem of the algorithm optimizes the growth rate of the cells and its inner problem results in the maximum compatibility between gene expression data and flux distribution in the metabolic network. In order to establish the maximum compatibility, a new soft constraint was introduced that compares the expression of the metabolic genes between and within the cells and tries to adjust the flux through a reaction based on the expression of its corresponding gene(s). In fact, the algorithm is capable to simulate the growth of multiple cells with considering the differences between their gene expression profiles. This algorithm was used to simulate the growth of cisplatin sensitive and resistant ovarian cancerous cells in the RPMI-1640 medium. The results show that the growth and lactate secretion rates of resistant cells are more than sensitive cells. These results are quantitatively comparable with experimental data. Establishment of the maximum compatibility using the inner linear problem was enabled the model to correctly predict lactate production due to the Warburg effect for both cell lines.

Key words: Genome-scale metabolic model, integration algorithm, microarray data, bi-level optimization.