A metaheuristic multi-objective optimization method for dynamical network biomarker identification as pre-disease stage signal

Abstract

Deciphering the signals that are attached to the transition from normal to disease stage is crucial in preventive medicine to understand the progression of complex diseases. Between normal and disease stages there exists the pre-disease stage, in which the disease is yet reversible towards the normal stage. Traditionally, molecular biomarkers have been used to identify the pre-disease stage. However, they have limitations because they have an individual and static nature. In complex diseases, the dynamics and interplays of certain genes have to be taken into account in order to identify the predisease stage. Therefore, in complex diseases, it is necessary to use dynamical network biomarkers (DNBs). The development of time-course omics data has been crucial to the use of DNBs as biomarker. In this article, a new two-step method is proposed for the identification of DNBs as pre-disease stage signal. In the first step, the relevant genes in the dataset are pre-filtered using a differential gene expression analysis. In the second step, the DNBs are identified, from a multi-objective optimization viewpoint, by using an Artificial Bee Colony based on Dominance (ABCD) algorithm. Specifically, identified DNBs optimize three objectives: they are the smallest gene network that shows the strongest signal in the earliest time-point of the disease progression and best correlates with the disease phenotype. The proposed method has been evaluated with five time-course microarray datasets and the results have been compared with five methods from other authors, surpassing their results. The effectiveness of the proposed method has been also proved with a leave-one-out cross-validation and a Gene Ontology term enrichment. In fact, the proposed method obtains values around 90% for accuracy, precision, recall, and F1 scores.