Abstract
Currently, cancer is increasingly becoming one of the major threats to human health. It is important to precisely identify gene combinations associated with cancers. Therefore, a dissolving P system based on membrane computing theory is proposed in this study to detect effective gene combinations related to cancer from high-dimensional micro-array data. We employ three classification functions with penalty terms to form a multi-objective optimization problem to further guide the gene combination selection process. The proposed P system is capable of helping to balance the global exploration and local exploitation abilities, and the penalty term is used to screen out redundant genes by reducing the number of genes automatically in each selected gene combination. Taking two micro-array datasets of prostate cancer as examples, the experimental results show that our method can identify lower-order gene combinations with higher classification accuracy than existing algorithms. We further visualize the classification effect on two datasets comprehensively and validate the selected representative genes biologically. The proposed method successfully achieves the gene combination selection task. Moreover, we comprehensively investigate the performance of our method on three other benchmark datasets and obtain satisfactory results.