Department of Aquaculture, Faculty of Agriculture, Universiti Putra Malaysia, 43400, Selangor, Malaysia.
Environ Sci Pollut Res Int. 2013 Mar;20(3):1586-95. doi: 10.1007/s11356-012-1027-5. Epub 2012 Jul 1.
This study represents a first attempt at applying a fuzzy inference system (FIS) and an adaptive neuro-fuzzy inference system (ANFIS) to the field of aquatic biomonitoring for classification of the dosage and time of benzo[a]pyrene (BaP) injection through selected biomarkers in African catfish (Clarias gariepinus). Fish were injected either intramuscularly (i.m.) or intraperitoneally (i.p.) with BaP. Hepatic glutathione S-transferase (GST) activities, relative visceral fat weights (LSI), and four biliary fluorescent aromatic compounds (FACs) concentrations were used as the inputs in the modeling study. Contradictory rules in FIS and ANFIS models appeared after conversion of bioassay results into human language (rule-based system). A "data trimming" approach was proposed to eliminate the conflicts prior to fuzzification. However, the model produced was relevant only to relatively low exposures to BaP, especially through the i.m. route of exposure. Furthermore, sensitivity analysis was unable to raise the classification rate to an acceptable level. In conclusion, FIS and ANFIS models have limited applications in the field of fish biomarker studies.
本研究首次尝试将模糊推理系统（FIS）和自适应神经模糊推理系统（ANFIS）应用于水生生物监测领域，通过选择非洲鲶鱼（Clarias gariepinus）中的生物标志物对苯并[a]芘（BaP）注射剂量和时间进行分类。鱼通过肌肉内（i.m.）或腹腔内（i.p.）注射 BaP。肝谷胱甘肽 S-转移酶（GST）活性、相对内脏脂肪重量（LSI）和四种胆汁荧光芳香族化合物（FACs）浓度用作建模研究的输入。在将生物测定结果转化为人类语言（基于规则的系统）后，FIS 和 ANFIS 模型出现了矛盾规则。提出了一种“数据修剪”方法来消除模糊化之前的冲突。然而，所产生的模型仅适用于相对较低水平的 BaP 暴露，尤其是通过 i.m.暴露途径。此外，敏感性分析无法将分类率提高到可接受的水平。总之，FIS 和 ANFIS 模型在鱼类生物标志物研究领域的应用有限。
