DL: Towards the Digital Salmon: From a reactive to a pre-emptive research strategy in aquaculture (DigiSal)

Salmon farming in the future must navigate conflicting and changing demands of sustainability, shifting feed prices, disease, and product quality. The industry needs to develop a flexible, integrated basis of knowledge for rapid response to new challenges. The DigiSal project has laid the foundations for a Digital Salmon: an ensemble of mathematical descriptions of salmon physiology. The project combines mathematics, high-dimensional data analysis, computer science, and measurement technology with genomics and experimental biology into a concerted whole. As the first application within our long-term vision, we chose challenges related to novel feedstuffs. Salmon are carnivores but today aquaculture provides more than half their fat and protein from plants, challenging the metabolic system and affecting fish health and the nutritional value of salmon meat. DigiSal has produced a series of constraint-based models of salmon metabolism in various tissues, life stages, and lipid nutrition states. They capture key metabolic processes converting feed into fillet and relate to available omics data on gene expression and metabolite amounts, though many vital processes (feeding behaviour, energy allocation, developmental biology, reproduction) are not yet integrated into the emerging modelling framework. By publishing the first metabolic model of a production animal, we are bridging the gap between production and systems biology and initiating a framework for adapting salmon breeding and nutrition strategies to modern feeds. Explicitly connecting metabolites, reactions, and genes, the SALARECON model links genome to metabolism and growth and can be tuned to specific genetic and environmental contexts by integration of domain knowledge and experimental data. Thus, SALARECON forms a transdisciplinary framework for diverse disciplines and data sets involved in salmon research and aquaculture. Tools developed for CBM of microbes and well-studied plants and animals can now be applied in production biology, providing a sharper lens through which to interpret omics data by requiring consistency with flux balances and other known constraints. This enables clearer analysis than classical multivariate statistics, which does not incorporate mechanistic knowledge. The SimSaLipiM model (due for 2024) elaborates on lipid metabolism, which accounts for most of the growth energy, and prototypes feed optimization combining economical and sustainability criteria. This points towards the ability to compose serviceable feeds under changing prices while minimizing the environmental impacts of feedstuffs. We identified a novel strain of Mycoplasma that dominates in the gut of farmed salmon at sea, but failed to culture it. Aiming to model the host-bacteria interplay in the salmonid gut, we collaborated with veterinarians studying gut health using cell culture of rainbow trout (salmon gut cell cultures did not yet exist). Combining gene expression data (RNA and protein), metabolomics and cell growth, we evaluate the realism of cell culture in the usual flasks vs. transwell culture which preserves the directionality of the gut epithelium. This metabolic modelling offers a relevant, consistent summary and basis for comparison of metabolic function in wet-lab model systems, and extends DigiSal's scope to a related farmed species. Modelling collaboration with the university of Minnesota suggests that Mycoplasma is tightly associated with its host, probably intracellular. Studies of omega-3 metabolism on different diets and life stages showed that salmon rearrange their fat metabolism before going to sea. Pioneering gene editing has verified the role of key genes in omega-3 metabolism, but active steering towards higher omega-3 in meat awaits results from exploration of similar genes in the more ecologically diverse brown trout. The metabolic models described above are the first of their kind for a production animal, integrating vast biological knowledge through machine-navigable open standards and facilitating analysis and dialogue with experts. Throughout its life, DigiSal has pushed for standards and best practices in systems biology, keeping data and models FAIR: Findable, accessible, interoperable, and reusable. We were a "star use case" of FAIRDOM, a multinational European foundation fostering FAIR data sharing. What DigiSal has started will be carried forward by Digital Life Norway's pilot project "Non-commercial business models for data driven ideas", which explores how to reconcile the values of open science with those of commercial actors. Specifically, the pilot supports the Ard technology transfer office in developing a business model to sustain modelling-based research output such as that from DigiSal, in dialogue with NCE Seafood Innovation Cluster, whose Industry insight report on "data sharing in the Norwegian aquaculture industry" represents broad consensus on data sharing for environmental, social, and governance purposes.

DigiSal has added metabolic modelling to the toolbox of NMBU's research in systems biology, genomics and nutritional physiology, in close interplay with the breeding company AquaGen. The models will serve as lenses for viewing future omics data, and will be regularly refined and extended, integrating and consolidating new knowledge. Our regulatory mapping of the salmon genome enriches the salmonid evolutionary genomics research at Cigene in particular, and salmon research and development globally. We are now engaging with life-cycle assessment research at NMBU based on our prototyping of feed optimization combining economical and sustainability criteria. NTNU will make use of its efficient metabolomics protocols in years to come, and has extended its gene editing expertise through collaboration with IMR. Ard innovation has used DigiSal as an inroad to value generation from data analysis and systems biology modelling, developing a use-case for the potential to integrate data and bring actionable insights on actual biology. DigiSal has served as an example of FAIR data and model management using open standards and interoperable format, highlighted in FAIRDOM's funding review, in several workshops with Digital Life Norway, in experience sharing with many labs across NMBU, and in dialogue with industry through the Seafood Innovation Cluster and NMBU's Green Data Lab. The Memote metabolic model testing framework is now routinely applied on models published and in development, and we were among the initiative-takers and continue to push towards more biologically meaningful model testing. In the long term, we believe aquaculture operations, research and development will adopt a hybrid of mechanistic modelling, machine learning and artificial intelligence which combines biological insight and data crunching in an efficient manner. This includes best practices of documentation, verifiability and reusability. We have blazed a trail of common models, standards and a shared understanding of value generation from model analysis in production biology. Our combination of in vitro assays and mathematical models providing a systems understanding of the salmon metabolism will in time allow the prediction of the physiological consequences of various feed recipes with a minimum of experimental effort. In the face of new challenges, the industry will be able to quickly reanalyse existing data and identify knowledge gaps, and to rapidly acquire the required data and place it into context of what we already have and know. Thus, we shift from a reactive to a pre-emptive research and development strategy.

Salmon farming in the future must navigate conflicting and shifting demands of sustainability, shifting feed prices, disease, and product quality. The industry needs to develop a flexible, integrated basis of knowledge for rapid response to new challenges. Project DigiSal will lay the foundations for a Digital Salmon: an ensemble of mathematical descriptions of salmon physiology, combining mathematics, high-dimensional data analysis, computer science and measurement technology with genomics and experimental biology into a concerted whole. DigiSal will focus on challenges of novel feedstuffs. Salmon are carnivores but today aquaculture provides more than half their fat and protein from plants, challenging the metabolic system and affecting fish health and nutritional value of salmon meat. The newly sequenced salmon genome and related resources will enable a tightly integrated theoretical-experimental study of mechanistic interactions among genetic and feed factors. Systems-oriented mathematical and statistical modelling will be central, using existing and novel knowledge e.g. on metabolic reaction networks to guide design of experiments through multiple iterations. Metabolic function of fish will be characterized via multiple omics technologies in feeding trials and in vitro tissue-slice culture. Gut microbiota will receive particular attention. The resulting massive data will be summarized via multivariate models to deliver a predictive understanding of a whole range of possible diets, much more efficiently than by traditional feeding trials alone. Data and models will be annotated using bio-relevant ontologies, so that new knowledge automatically connects to that which already exists. Future challenges will be met by quickly reanalysing existing information and understanding of salmon biology, identifying knowledge gaps, acquiring new data and incorporating it into a unified whole. Thus, we begin a shift from a reactive to a pre-emptive R&D strategy in aquaculture.