Abstract This study shows that genetic algorithm (GA) is a suitable method for selecting wavelengths for partial least squares (PLS) calibration of mixtures with almost identical spectra without loss of prediction capacity employing spectrofluorimetric method. A training set of mixtures containing different concentrations of ofloxacin (OFL) and riboflavin (B2) were prepared to be used as calibration set to check the prediction ability of GA-PLS models due to spectral overlapping of these constituents. Each model was validated using a validation set and then real samples were analyzed. Linear calibration curves were obtained in the 0.5-5.0 and 2.0-10.0 µg mL-1 range for ofloxacin and riboflavin, respectively. To preprocess the data matrices, the orthogonal signal correction (OSC) was used and the analysis results were statistically compared. The methods accuracy for simultaneous determination of ofloxacin and riboflavin, were evaluated by the root mean square errors of prediction (RMSEP) which were 0.0868 and 0.158 for ofloxacin and riboflavin, respectively, and relative standard error of prediction (RSEP) which were 2.738 and 2.846 for ofloxacin and riboflavin, respectively using OSC-GA-PLS models. This procedure allows the simultaneous determination of OFL and B2 in human urine and serum samples with good reliability of the determination.

Resumen. En este estudio se muestra que los algoritmos genéticos (GA) son un método adecuado para seleccionar longitudes de onda para calibrar métodos de mínimos cuadrados parciales (PLS) de mezclas, con espectros casi idénticos sin perder la capacidad de predicción al utilizar el método espectrofluorimétrico. Se preparó un conjunto de entrenamiento de mezclas que contenían diferentes concentraciones de ofloxacina (OFL) y riboflavina (B2), que se usaron como un conjunto de calibración, para verificar la capacidad predictiva de los modelos GA-PLS debidos a la superposición espectral de estos componentes. Cada modelo se comprobó usando un conjunto de validación y, posteriormente, las muestras reales se analizaron. Se obtuvieron curvas de calibración lineal en los intervalos de 0.5-5.0 y 2.0-10.0 µg mL-1, para ofloxacina y riboflavina, respectivamente. Para el preprocesamiento de las matrices de datos, se utilizó la corrección ortogonal de señal (OSC) y los resultados del análisis se compararon estadísticamente. La precisión de los métodos para la determinación simultánea de ofloxacina y riboflavina se evaluó mediante la predicción de los errores cuadráticos medios (RMSEP) que fueron 0.0868 y 0.158 para ofloxacina y riboflavina, respectivamente, y la predicción del error relativo estándar (RSEP) que fueron 2.738 y 2.846 para ofloxacina y riboflavina, respectivamente, utilizando los modelos OSC-GA-PLS. Este procedimiento permite la determinación simultánea de OFL y B2 en orina humana y en muestras de suero con buena confianza en su determinación.

Dispersive liquid-liquid microextraction (DLLME) and spectrofluorimetry, coupled with chemometrics methods, are proposed in this work for the pre-concentration and determination of ofloxacin concentration in spiked human urine. Chloroform and acetonitrile were selected as the extraction and dispersive solvents by the one-variable-at-a-time process. The Box-Behnken design was used to optimize the other variables, including the volume of extraction and dispersion solvents, solution pH, and ionic strength. A linear calibration curve was obtained in the 5.0-120.0 ng mL-1 range under optimal conditions with a detection limit of 1.61 ng mL-1 and correlation coefficient of 0.9948. A relative standard deviation (RSD) of 1.13% was obtained for seven consecutive replicates. Parallel factor analysis (PARAFAC) and partial least square (PLS) modeling were applied for the multivariate calibration of the spectrofluorimetric data. To pre-process the data matrices and predict the model results, the orthogonal signal correction (OSC) was used, and the analysis results were statistically compared. The methods accuracy values for ofloxacin determination, evaluated by the root mean square errors of prediction (RMSEP) and relative standard error of prediction (RSEP), were 0.82 and 1.12 using OSC-PLS, and 0.31 and 0.42 using OSC-PARAFAC models, respectively. Ofloxacin can be reliably determined in human urine samples through the proposed procedure, according to the results.