MORE interpretable multi-omic regulatory networks to characterize phenotypes

Abstract

The emergence of omics has revolutionized how we study and understand biological systems, enabling breakthrough discoveries with profound implications for medicine, agriculture, biotechnology, and more. However, with the help of advanced computational tools and artificial intelligence, meaningful patterns and relationships can now be uncovered in omics data, offering a unique opportunity to gain a deeper understanding and contribute to new insights into the complex regulatory mechanisms of biological systems. In this context, we have developed MORE (Multi-Omics REgulation), a tool designed to identify relevant regulations of the gene expression for the biological system under study and subsequently construct the regulatory networks for the considered experimental conditions. The presented method not only allows the incorporation of prior biological information into network construction but also can infer relationships de novo in the absence of such information. Moreover, we effectively addressed multicollinearity issues inherent in such data types, ensuring precise and reliable inference of regulatory networks when performing GLM models. In our comparison to KiMONo, our tool exhibited superior evaluation metrics, including F1-score, R2, and computational efficiency. Finally, applying our tool to a real ovarian cancer dataset yielded intriguing and biologically meaningful results. Our developed methodology represents a versatile and powerful multi-omic regulatory network inference approach, demonstrating good performance and applicability to real-world biological datasets. It is freely available at https://github.com/ConesaLab/MORE.git.