BONOBO: Bayesian Optimized Sample-Specific Networks Obtained by Omics Data

Abstract

Correlation networks can provide important insights into biological systems by uncovering intricate interactions between genes and their molecular regulators. However, methods for estimating co-expression networks generally derive an aggregate population-specific network that represents the mean regulatory properties of the entire population and hence falls short in capturing heterogeneity across individuals. While numerous methods have been proposed to estimate sample-specific co-expression networks, they fail to estimate positive semidefinite correlation networks and, hence, are subject to misinterpretation. To fill this gap in co-expression network inference, we introduce BONOBO (Bayesian Optimized Networks Obtained By assimilating Omics data), a scalable Bayesian model for deriving individual sample-specific co-expression networks by acknowledging heterogeneity in molecular interactions across individuals. For each sample, BONOBO imposes a Gaussian distribution on the log-transformed, centered gene expression and a conjugate Inverse Wishart prior distribution on the sample-specific co-expression matrix constructed from assimilating all other samples in the data. BONOBO yields a closed-form solution for the posterior distribution of the sample-specific co-expression matrices by combining the sample-specific gene expression with the prior distribution. We demonstrate the advantages of BONOBO using several simulated and real datasets. BONOBO is computationally scalable and available as open-source software through the Network Zoo package (from netZooPy v0.10.0; netzoo.github.io). A preprint associated with this abstract can be found on bioRxiv (doi: 10.1101/2023.11.16.567119v1).

E. Saha and V. Fanfani—These authors contributed equally.