IEEE/ACM Trans Comput Biol Bioinform. 2018 Nov-Dec;15(6):2053-2059. doi: 10.1109/TCBB.2017.2771360. Epub 2017 Nov 9.
Gene Regulatory Network (GRN) is a virtual network in a cellular context of an organism, comprising a set of genes and their internal relationships to regulate protein production rate (gene expression level) of each other through coded proteins. Computational Reconstruction of GRN from gene expression data is a widely-applied research area. Recurrent Neural Network (RNN) is a useful modeling scheme for GRN reconstruction. In this research, the RNN formulation of GRN reconstruction having single objective function has been modified to incorporate a new objective function. An existing multi-objective meta-heuristic algorithm, called Archived Multi Objective Simulated Annealing (AMOSA), has been modified and applied to this bi-objective RNN formulation. Executing the resulting algorithm (called AMOSA-GRN) on a gene expression dataset, a collection (termed as Archive) of non-dominated GRNs has been obtained. Ensemble averaging has been applied on the archives, and obtained through a sequence of executions of AMOSA-GRN. Accuracy of GRNs in the averaged archive, with respect to gold standard GRN, varies in the range 0.875 - 1.0 (87.5 - 100 percent).
基因调控网络（GRN）是生物体细胞环境中的一个虚拟网络，由一组基因及其内部关系组成，通过编码蛋白相互调节彼此的蛋白质产生速率（基因表达水平）。从基因表达数据中计算重建 GRN 是一个广泛应用的研究领域。递归神经网络（RNN）是 GRN 重建的有用建模方案。在这项研究中，具有单个目标函数的 GRN 重建的 RNN 公式已被修改为包含新的目标函数。一种现有的多目标启发式算法，称为存档多目标模拟退火（AMOSA），已被修改并应用于这种双目标 RNN 公式。在基因表达数据集上执行由此产生的算法（称为 AMOSA-GRN），获得了一组非支配 GRN（称为存档）。对档案进行了集合平均，并通过 AMOSA-GRN 的一系列执行获得。与黄金标准 GRN 相比，平均档案中 GRN 的准确性在 0.875 到 1.0 之间变化（87.5%到 100%）。
