P System as a Computing Tool for Embedded Feature Selection and Classification Method for Microarray Cancer Data

TL;DR: In this article, a multi-objective binary particle swarm optimization (MObPSO) algorithm was proposed to select informative genes from microarray data, where the kernel P system (kP) was used as the variant of the P system to improve the performance of the intelligent algorithm.

Abstract: Selection of relevant genes is the crucial task for sample classification in microarray data, where researchers try to identify the smallest possible set of genes that can still achieve good predictive performance. Due to the problem of higher risk of overfitting in wrapper methods and sensitivity of the best embedded way to filter out factor that leads to unstable model and significantly different gene subsets, in this paper, we propose a novel model for evaluating and improving techniques for selecting informative genes from microarray data. This model inspired by membrane computing and used the kernel P system (kP) as the variant of the P system to improve the performance of the intelligent algorithm, multi-objective binary particle swarm optimization (MObPSO). The proposed model consists of two main parts. First, kP-MObPSO, which resembles a wrapper type feature selection, and the second part that improves the results of the first part through an embedded feature selection and classification idea based on the kP system. Division, rewriting, and input/output rules are used to make interaction among the genes inside and between the particles. The proposed model applied to the colorectal and breast dataset contains 100 genes with six attributes. The embedded part of the model extracts the marker gene sets indicate more stability and reliability based on ROC measure as well as better error rate in comparison to the wrapper part of the model. In the paper, the lowest error rate by an embedded model is displayed as 0.1111 for breast cancer and 0.0769 for colorectal data.