Oncogene inference optimization using constraint-based modelling incorporated with protein expression in normal and tumour tissues

Abstract

Cancer cells are known to exhibit unusual metabolic activity, and yet few metabolic cancer driver genes are known. Genetic alterations and epigenetic modifications of cancer cells result in the abnormal regulation of cellular metabolic pathways that are different when compared with normal cells. Such a metabolic reprogramming can be simulated using constraint-based modelling approaches towards predicting oncogenes. We introduced the tri-level optimization problem to use the metabolic reprogramming towards inferring oncogenes. The algorithm incorporated Recon 2.2 network with the Human Protein Atlas to reconstruct genome-scale metabolic network models of the tissue-specific cells at normal and cancer states, respectively. Such reconstructed models were applied to build the templates of the metabolic reprogramming between normal and cancer cell metabolism. The inference optimization problem was formulated to use the templates as a measure towards predicting oncogenes. The nested hybrid differential evolution algorithm was applied to solve the problem to overcome solving difficulty for transferring the inner optimization problem into the single one. Head and neck squamous cells were applied as a case study to evaluate the algorithm. We detected 13 of the top-ranked one-hit dysregulations and 17 of the top-ranked two-hit oncogenes with high similarity ratios to the templates. According to the literature survey, most inferred oncogenes are consistent with the observation in various tissues. Furthermore, the inferred oncogenes were highly connected with the TP53/AKT/IGF/MTOR signalling pathway through PTEN, which is one of the most frequently detected tumour suppressor genes in human cancer.

Optimization of a modeling platform to predict oncogenes from genome‐scale metabolic networks of non‐small‐cell lung cancers

“…However, duality transformation is difficult for multilevel optimization problems, such as the TLOP in this study. We applied the NHDE algorithm (see Supplementary File S2) , which has been used to solve oncogene inference problems [18,21] , to infer the oncogenes of LUAD and LUSC. The problem [Eq (1)] consisted of the crisp objectives and fuzzy equal objective to introduce a combination of weighted-sum and minimum decisions for evaluating the fitness,  D , which was used in the NHDE algorithm as follows:…”

“…obtained through MFVA, and were used to determine the trend of flux change between dysregulated case and normal situation in terms of seven categories of classification [21].…”

“…In this study, we first applied the CORDA method [30] was used to analyze the reactions, metabolites, and enzyme-encoding genes of these GSMNs to discriminate differential expressions between normal and cancer cells. The oncogene inference optimization formulation [18,21] was used to mimic gene screening procedures in a wet lab to evaluate the mechanism by which gene dysregulations induce tumorigenesis.…”

“…Genetic alterations and epigenetic modifications of cancer cells result in the abnormal regulation of cellular metabolic pathways that differ from normal cells. GSMNs combined with constraint‐based modeling approaches can predict the metabolic reprogramming of cancer cells to reveal oncogenes, essential enzymes, and drug targets for developing novel medical treatments [ 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 ]. The first large‐scale reconstructed metabolic model for cancer was built based on the gene expression data of all cancer cell lines in the NCI‐60 collection and the human general metabolic network (Recon 1) [ 14 , 23 ].…”

“…Recon 2.2 and Recon 3D are the most comprehensive human genome‐scale network reconstructions [ 25 , 26 ]. Recon 2.2 was incorporated with the Human Protein Atlas (HPA) [ 27 ] to reconstruct GSMNs of the colorectal tissue [ 18 ] and head‐and‐neck tissue [ 21 ] at normal and cancerous states. Recon 2M.2 [ 28 , 29 ] integrated with RNA‐Seq data from NCI‐60 cell lines presented a systematic framework for the generation of gene–transcript–protein–reaction that enables the accurate prediction of metabolic behaviors.…”