A BAYESIAN MULTI-OMICS STRATEGY TO INVESTIGATE HOST–MICROBIOTA DYNAMICS IN AQUACULTURE

Introduction

Modern aquaculture increasingly requires analytical frameworks capable of capturing the complexity of host responses to environmental and nutritional challenges. Isolated measurements often provide only partial insight, whereas integrative approaches can reveal the coordinated behavior of host physiology, microbial communities, and environmental drivers (Ruiz-Perez et al., 2021). The study of microbiota in aquaculture has traditionally focused on describing taxonomic composition, often across different body compartments such as gut, skin, gills, and the surrounding water. However, these datasets are rarely interpreted as parts of a connected biological system. Therefore, integrating microbial profiles from distinct host-associated niches and linking them with host-centered omics, such as transcriptomics, creates the opportunity to move from descriptive ecology toward mechanistic interpretation. This systems-level perspective is especially relevant in aquaculture, where performance traits, welfare, disease resistance, and environmental tolerance are shaped by multiple interacting biological layers. Among the available approaches, Bayesian networks (BN) are particularly attractive because they allow the inference of probabilistic relationships and within complex biological systems and support inferential hypothesis generation (Moroni et al., 2025). Their application in aquaculture could therefore provide a powerful route to uncover the host-microbe cross-talk and to support the development of predictive and function-based models of animal response.

Materials and methods

This study adopted an integrative multi-experiment approach. For this purpose, different trials involving dietary inclusion of functional ingredients, environmental contaminants, and alternative feed formulations were considered. Microbiota datasets derived from different compartments (gut, skin, water) and where available, they were further combined with additional omics layers (host gene expression data). The base of this study was the integration with the Bayesian network inferences, which were performed using the SAMBA, a computational platform developed by our research group (Soriano et al., 2023). This approach allowed the reconstruction of probabilistic causal networks describing dependencies between microbial features and omics data. Network structures were then processed using a customized R-based pipeline composed by a suite of network metrics, including degree centrality, weighted connectivity, betweenness, closeness, and hierarchical properties such as ancestor–descendant relationships and node depth. The strength and direction of inferred causal relationships were extracted and analysed from the conditional probability tables (CPTs), associated with each node. All these values were then integrated into matrices to compute total causal effects, capturing both direct and indirect propagation of influence across the network. Based on these outputs, nodes were classified into functional roles (e.g., initiators, drivers, propagators, mediators, and responders), reflecting their behaviour, position, and influence within the microbial system.

Results and discussion

The results obtained so far demonstrate the potential of this pipeline to integrate 16S rRNA gene-based microbiota datasets and multi-omics data through Bayesian network modeling. This approach made it possible to explore the potentially causal relationships within the microbial communities, such as those established between fish skin and surrounding water, across time. In particular, the analysis highlighted how bacterial taxa shared by both compartments change over time and how their ecological roles within the community may shift in response to the experimental stimulus (Figure 1a). At the multi-omics level, the downstream integration of metataxonomic and gene expression data enabled the characterization of ecological roles using both topological and non-topological metrics from BN, leading to the identification of mixed clusters composed of bacterial taxa and host genes, which responded to the experimental trial (Figure 1b). These findings represent a valuable starting point for a more detailed functional characterization and provide new insight into the processes underlying host-microbiota cross-talk. Future developments of this pipeline should further increase the level of omics integration, including the identification of key receptors potentially involved in host-bacterial interactions and the incorporation of metabolomic profiles to complement and strengthen the interpretation of these mechanisms. In the long term, this framework may contribute to a better understanding the details of how animals respond to experimental perturbations and may support the development of mitigation strategies, combining microbiome- and host-based solutions, aimed at improving welfare and productivity in aquaculture under different scenarios.

Figure 1. Important taxa ranked by Betweenness (a). PCA and radar plot of the roles within BN (b).

Funding

This study was funded by BreamHOLOBIONT (PID2023-146990OB-I00), AICO 2025 (CIAICO/2024/281) and APOSTD 2024 (CIAPOS/2024/092).

References

Ruiz-Perez, D., et al. (2021) Dynamic Bayesian Networks for Integrating Multi-omics Time Series Microbiome Data. mSystems 6:10.1128/msystems.01105-20.

Moroni, F., et al. (2025) One Function, Many Faces: Functional Convergence in the Gut Microbiomes of European Marine and Freshwater Fish Unveiled by Bayesian Network Meta-Analysis. Animals 15, 2885.

Soriano, B., et al. (2023) SAMBA: Structure-Learning of Aquaculture Microbiomes Using a Bayesian Approach. Genes 14, 1650.