Comparison between Univariate and multivariate methods in Regression analysis of Sensor arrays

Array-based sensor platforms are inspired by the mammalian olfactory system. Multiple sensor elements in these devices selectively interact with target analytes, producing a distinct pattern of response and enabling analyte identification. It provides a multidimensional data set that needs to be processed by multivariate analysis methods. In quantitative analyses, the relation of sensor array responses and different concentration of analyte can be obtained by univariate and multivariate regression methods. In the univariate approach, response vector of each concentration convert to a value by calculating the Euclidean norm. While in multivariate regression methods, the relationship is built between the sensor array responses and analyte concentration. Many times these two methods used interchangeably in analyzing the sensor arrays data [1]. However, it is a challenge that univariate and multivariate methods can quantifying the species in complex mixture the same or not. To compare two univariate and multivariate method in analysis the sensor arrays, the operation of a sensor array based strip in four different real matrix (cell culture, milk, , Orange juice and tap water) were considered to evaluate the dependency of color values (R, G and B) on the concentration (pH values) [2]. Since the studied real samples may be complex, the standard addition method was applied for pH determination. To do so, for each real sample solution, the response of the strip was measured by dipping it in the sample solution. The Euclidean norm and PLS calibration models were built for each real sample systems. To evaluate the ability of the methods, the correlation coefficient (R2) between predicted and the actual pHs calculated for multivariate methods and between the Euclidian norm and pHs for univariate method was compared. Also, the models were used to predict the pH of unknown sample and compare by the obtained pH of unknown samples calculated by Euclidean norm method. However, the R2 obtained from PLS model and Euclidean norm methods are so close to each other, (cell culture: 0.97, 0.99; milk: 0.94, 0.96; Orange juice: 0.97, 0.98; and tap water: 0.97, 0.97, related to PLS model and Euclidean norm, respectively.), but their ability in prediction the pH of unknown sample are significantly different (Recovery for cell culture: 0.98.9, 103.5; milk: 90.5, 80.93; Orange juice: 91.8, 31.8; and tap water: 104.4, 81.2, related to PLS model and Euclidean norm, respectively). It can be concluded that chemometrics method can be a better candidate for prediction the unknown concentration in sensor arrays data.