Multi-objective consensus optimization for gene regulatory networks inference: A preference-based approach

Abstract

Gene regulatory networks (GRNs) model key gene interactions, enabling the understanding of essential biological processes and their relationship with diseases. Inferring GRNs from expression data is fundamental in computational biology. However, existing methods exhibit limitations like domain biases and a lack of biological knowledge integration that affect their performance in in-vivo experimentation, particularly when several conflicting objectives are considered. To address these challenges, we propose a new approach that adopts a preference-guide selection mechanism aimed at helping the partitioner direct the search towards regions of high biological relevance by defining reference points in the objective space. This mechanism is integrated into MO-GENECI, a multi-objective evolutionary algorithm designed to optimize consensus between multiple machine learning techniques through biologically relevant objectives. Driven by research questions, the proposed approach is evaluated on 43 GRNs from benchmarks like DREAM3 and DREAM4, and real-world databases such as TFLink, using AUROC and AUPR metrics. The results demonstrate that the generated consensus networks obtained by using the preference selection outperform the original algorithm in quality and accuracy and reduce computational effort, especially in large networks. PBEvoGen achieved mean AUROC and AUPR values of 0.67 and 0.23 across 43 benchmark networks, improving the already state-of-the-art MO-GENECI by 1.2% and 4.3%, respectively. This combination of expert knowledge and evolutionary algorithms offers a robust, efficient methodology for GRN inference. The source code is hosted in a public repository at GitHub under MIT license: https://github.com/AdrianSeguraOrtiz/PBEvoGen. Moreover, to facilitate its installation and use, the software associated with this implementation has been encapsulated in a Python package available at PyPI: https://pypi.org/project/geneci/2.5.1