Quantitative structure-property relationship study of spectral properties of green fluorescent protein with support vector machine

Green fluorescent protein (GFP) is an autofluorescent protein that has been widely used in the biomedical sciences for molecular imaging applications. Computational approach for predicting the spectral properties of GFP offers great benefit for the design and engineering of novel color variants. Herein, we present a quantitative structure-property relationship (QSPR) study to model the spectral properties (e.g. excitation and emission maxima) of GFP chromophores using support vector machine (SVM). The data set is composed of 19 chromophores from GFP color variants and 29 synthetic GFP chromophores based on the imidazolinone scaffold. Quantum chemical descriptors were used to provide information on the physicochemical properties of the chromophores. Such descriptors were mapped onto a higher dimensional space via kernel functions (e.g. linear, polynomial and radial basis function kernels) and learning is then performed using SVM. The predicted spectral properties were well correlated with their experimental values as observed from correlation coefficient in the range of r = 0.953-0.979. Predictive performance of excitation maxima (r = 0.967-0.979) outperformed that of the emission maxima (r = 0.953-0.961). The present strategy holds great promise for expanding the spectral repertoire of GFP by facilitating the rational design of novel color variants.