Estimating the body composition of fish is an important step in fish nutrition, as it is the basis to calculate nutrient retention efficiencies and utilization rates. In fish nutrition studies, nutrient retention can be used to assess the nutritional quality of diets, but also to support the definition of nutrient requirements and the estimation of nutrient waste outputs. From the point of view of fish farmers, knowing the body composition of fish more regularly throughout the production cycle can help in the search for more effective feeding strategies and also to have a better quality-control of their products.
Usually, body composition of fish is estimated through analytical methods, such as the ones described by the Association of Official Analytical Chemists (AOAC). Despite being reliable and robust methods to estimate the body composition of fish, these analytical methods are not always a viable option to estimate the body composition of fish, since they are time-consuming and expensive. This translates into a limitation of the number of samples collected for body composition analysis, which can hinder a detailed analysis of nutrient flux in fish.
Here we present ficoEst, a public web-tool for researchers and fish farmers to estimate the body composition of farmed fish (https://webtools.sparos.pt/ficoest/). ficoEst uses calibrated and validated mathematical models to provide estimates on the body composition of different fish species (i.e., gilthead seabream, European seabass, meagre, rainbow trout, Atlantic salmon and Nile tilapia). ficoEst can be seen as a complementary tool to support studies on fish nutrition or to increase information collected at the farm level, whenever analytical methods are not a viable option.
Data collection and model development
Data on the whole-body composition of fish were collected for the abovementioned fish species. All data was processed into a standard format and analyzed. The outcomes of the data analysis process were crucial in providing a solid foundation for the model development phase, e.g., for identifying the key explanatory variables.
Different types of models and calibration methods were developed and tested, aiming to select the best methods in estimating the body composition of fish. All developed models fall into one of the following three families:
To select the best combination of model and calibration method (e.g., least squares, Huber loss minimization, mixed-effects, seemingly unrelated regressions), per family, we used cross-validation, where different error metrics were evaluated (e.g., MAPE, RMSE, AE). After selecting the model and calibration method to use per family, the models were calibrated and validated for each species. Figure 1 shows the model validation results for the BC3 model (similar analysis was conducted or BC1 and BC2 models, but not presented here).
All modules that compose the ficoEst tool were implemented in R. The back-end engine consists of different functions used to: compile, process, and plot data; evaluate, calibrate, validate and run models; and generate reports. The front-end (i.e., user-interface) consists of functions used to create input controls, and informational and navigational elements. Figure 2 shows the user-interface. To get estimates on the crude protein, crude lipids, water, ash, phosphorus and energy content of fish, users need to enter data on the body weight, and/or water and ash content of fish (depending on the model selected).
This work is part of project 47175_FICA, supported by Portugal and the European Union through FEDER/ERDF, COMPETE 2020 and CRESC Algarve 2020, in the framework of Portugal 2020.