Exploring the structural basis of organic compounds by predicting experimental IR peaks: a machine learning analysis

To understand the structural foundation of organic compounds is crucial in fields like chemistry and materials science. This study is a machine learning quest for predicting the experimental carbonyl peaks in the infrared (IR) spectrum of organic compounds from Modred and RDKit descriptors. The results show that FractionCSP3 is the most correlating descriptor for both types of descriptors. The Extra Trees (ET) regression yields the best performance with its coefficient of determination (R²) of 0.72-0.78. The analysis of models with SHapley Additive exPlanations (SHAP) shows that BCUT2D_MRLOW (RDKit) and FCSP3 (Modred) are the most influential descriptors. Hyperparameter tuning with 50 estimators optimizes model performance. Additionally, the calculated synthetic accessibility (SA) scores have 0.00-0.15 range to provide insights into the feasibility of synthesis. The current findings demonstrate the power of machine learning in uncovering the structural basis of organic compounds and predicting their experimental IR peaks.