QSPR study on the maximum absorption wavelength of Indoline dyes

The quantitative structure-property relationship (QSPR) for the maximum absorption wavelength (λ_max) of indoline derivatives was investigated. Two hundred and eight molecular descriptors were calculated by using RDKit software, and four key ones were screened through stepwise regression (SR) method. Multiple linear regression (MLR) and support vector machine (SVM) regression were used to develop the calibration model respectively. The developed models were assessed with leave-one-out cross-validation and external test set validation. The prediction statistics of the established MLR model were R_test²=0.5925, Q_F3,test²=0.5451, CCC_test=0.7725, r¯_m,test²=0.4298, ∆r_m,test²=0.2769, Q_LOO²=0.4759, r¯_m,LOO²=0.3875, and ∆r_m,LOO²=0.3213. The prediction statistics of the established SVM model were R_test²=0.8376, Q_F3,test²=0.8186, CCC_test=0.9021, r¯_m,test²=0.7345, ∆r_m,test²=0.0620, Q_LOO²=0.7493, r¯_m,LOO²=0.6505, and ∆r_m,LOO²=0.1562. Obviously, the accuracy and robustness of SVM model are higher than these of MLR model. All leverage values (h_i) for both models are below the critical threshold (h^⁎ = 0.29), indicating that all samples fall within the applicability domain of the models. The analytical result of property cliffs demonstrates the property cliffs at a low level, which does not mess up the developed models. It is demonstrated that the proposed method is practicable for predicting and studying the λ_max of indoline derivatives. Hence, Shapley additive explanations (SHAP) analysis was applied to interpret the SVM model. The result of SHAP analysis demonstrates that the descriptor of VSA_EState4 has the largest influence on the value of λ_max, indicating that the introduction of thiophene rings enhances electron delocalization in the conjugated system, thereby promoting a red shift in λ_max.