A study of boosting molecular descriptors with quantum-derived features in prediction of maximum emission wavelengths of chromophores
Bartłomiej Fliszkiewicz · Chemical Data Collections · 2021
The following research assesses the possibility to predict maximum emission wavelenghts of organic compounds based on quantum chemistry properties of the compounds’ fragments. The predictions are compared with predictions based on molecular descriptors and fingerprints widely applied in cheminformatics. Machine learning was applied in an effort to establish relationship between structures of organic compounds and their optical properties. Multilinear, gradient boosting and random forest regressions were chosen into the studies. Chosen validation metrics revealed that traditional descriptors provide better results than the fragments based quantum-derived features.