QSPR Analysis of Infant Infectious Diseases Using Machine Learning

Abstract

This study utilizes the neighborhood topological indices to assess the predictive performance of three machine learning models — Linear Regression (LR), k-Nearest Neighbors (KNN), and Random Forest (RF) — to model the relationships between the topological indices and some physicochemical properties (PHPs) of drugs used to treat infant infectious diseases. Root Mean Square Error (RMSE) is used as the evaluation metric across properties such as : boiling point (Bp), molar reactivity (Mr), molar volume (Mv), polarity (P), enthalpy of vaporization (Ev), melting point (Mp), and surface tension (St). The results indicate that kNN consistently provides the best predictive performance, particularly in cases involving complex relationships. LR performs well in linear cases but struggles with non-linear data. RF, although demonstrating strength in isolated cases, tends to exhibit the highest RMSE values. Additionally, the analysis of the Neighborhood Topological Indices (NTIs) shows that NRRG performs better in predicting properties like Bp, Ev, and P, whereas indices such as NFG, NRG, NM1, NSCI, NGA, and NM2 yield more accurate predictions for Mp, St, Mv, Mr, and Fp, respectively. These findings emphasize the importance of selecting appropriate models and indices in QSPR analysis.