Membrane Computing to Model Feature Selection of Microarray Cancer Data

Cancer is the main public health issue in most places of the world due to the difficulties in its early diagnosis and to begin early treatment. In the recent years many techniques have been proposed to tackle the high dimensionality in cancer datasets. This paper proposed a membrane-inspired feature selection method to utilize the potentials of membrane computing features such as decentralization, non-determinism, and maximal parallel computing for feature selection of cancer data. Kernel p system- one of the variants of membrane computing- is defined based on multi objective binary particle swarm optimization feature selection method through nine steps involving definitions of objects, compartments, rules and output. Matlab software is used to model the proposed approach. The proposed model evaluated by cell line data of breast cancer with six samples of papillary infiltrating ductal carcinoma and Carcinosarcoma disease state. As the initial attempt to come out with a model, division rule and sequential computation on Matlab are used as tools to define potential of membrane computing in trading space against time computation. The evaluation results indicate the proposed model based on kernel p system computation with distributed compartments is capable in distinguishing marker genes.