A multi-objective based PSO approach for inferring pathway activity utilizing protein interactions

The pathway information of a given microarray gene expression data can be collected from the available public databases. Inferring the activity of a pathway is a crucial task in functional genomics. In general, the set of genes that are associated with a given pathway are equally considered for measuring goodness. But the contribution of each gene should be quantified differently. In the current study, we have quantified the degrees of relevance of different genes participating in a pathway by optimizing different goodness measures of pathway activity. Two popular goodness measures, namely t-score and z-score are modified to measure the goodness of the weighted gene vectors. Moreover, another goodness measure based on the protein-protein interaction scores of pairs of genes participated in a pathway is utilized as another objective function. All these measures are designed to handle the weighted importance of individual genes. The search capability of a multiobjective based particle swarm optimization (PSO) is utilized for searching the appropriate relevance vectors for different genes. The proposed approach is applied to five real-life gene expression datasets, and the performance is compared with eight existing feature selection methods. The comparative results demonstrate the superiority of the proposed particle swarm optimization based technique. The efficacy of the performance of the proposed method is validated by using a statistical significance test, and further, a biological significant test is done to justify the biological relevance of the extracted pathway-based gene markers.