SelectWave: A graphical user interface for wavelength selection and spectral data analysis

Abstract

Studies on the determination of chemical compounds with different spectroscopy devices have become a hot topic in the scientific literature. A wide variety of programs are used to develop quantitative calibration models via different chemometric techniques in the scientific studies. However, there is a limited number of free and user-friendly software for creating quantitative determination models based on multivariate data analyses. In this study, we aimed to transform functions from R packages, which are widely used in spectral data analysis, into a free application accessible via the web. The application (SelectWave) has four sub-menus including data input, pre-analysis, post-analysis and about tabs. Data for dependent and independent variables should be loaded as calibration and validation sets separately. The pre-analysis tab includes four derivative functions and five pretreatment algorithms for spectral data analysis. Wavelength selection is possible with filter methods Variable Importance on Projections (VIP), Selectivity Ratio (SR), significance Multivariate Correlation (sMC) and minimum Redundancy Maximum Relevance (mRMR), and PLS based wrapper methods Interval Partial Least Squares (iPLS), Genetic Algorithm (GA), Iterative Predictor Weighing (IPW) and Uninformative Variable Elimination (UVE) under the post analysis tab. During the data modeling phase, the application provides Partial Least Squares Regression (PLSR) and Support Vector Machines (SVM) regression to the user. External validation can be performed using separate test set data. After the modeling process, evaluation statistics can be seen on the screen and automatically saved as a csv file under the user's working directory. The results of the variable selection can be inspected visually in the user interface. The developed application was tested on a personal computer (Intel Core i3, 4 GB RAM, ×64 processor, Microsoft Windows 10 Home) using spectral data from amylopectin analyses of maize flour samples (n ​= ​200) and on a more powerful computer using various data sets. The application is aimed at researchers who want to develop a multivariate quantitative calibration model with data obtained from any spectral device.