Nuclear abnormalities of marine fish erythrocytes.

The aim of this study was to monitor erythrocyte nuclear abnormalities (NA) including micronuclei (MN) in cultured and wild sea bass Dicentrarchus labrax and wild mullet Mugil spp. Seasonal sampling was performed at seven locations along the eastern coast of the Adriatic Sea. The frequency of NA and MN was positively correlated to temperature (NA: P < 0.05, r = 0.11; MN P < 0.05, r = 0.10), and there was also a positive correlation between NA and MN frequency (P < 0.001, r = 0.43). The lowest NA and MN values for both fish species were recorded in spring, while the highest were recorded in autumn. Significantly higher frequency of NA was seen in D. labrax compared to Mugil spp., while MN frequency was low in both species and not significantly different. There was no significant difference in NA and MN frequency between cultured and wild D. labrax sampled in the same month, and there was no difference between wild Mugil spp. sampled near or far from fish farms. In view of sampling sites, the highest values were detected in fishes from the Limski Channel, the lowest from the Janjina location.

Classification accuracy of algorithms for blood chemistry data for three aquaculture-affected marine fish species.

The objective of this study was determination and discrimination of biochemical data among three aquaculture-affected marine fish species (sea bass, Dicentrarchus labrax; sea bream, Sparus aurata L., and mullet, Mugil spp.) based on machine-learning methods. The approach relying on machine-learning methods gives more usable classification solutions and provides better insight into the collected data. So far, these new methods have been applied to the problem of discrimination of blood chemistry data with respect to season and feed of a single species. This is the first time these classification algorithms have been used as a framework for rapid differentiation among three fish species. Among the machine-learning methods used, decision trees provided the clearest model, which correctly classified 210 samples or 85.71%, and incorrectly classified 35 samples or 14.29% and clearly identified three investigated species from their biochemical traits.