Abstract
Quantitative structure-property relationship (QSPR) studies rely on molecular descriptors that link chemical structure to physicochemical properties. However, the rapid proliferation of topological indices has led to many with limited predictive utility. To address this, we introduce a new class of temperature-based spectral topological indices derived from eigenvalues of temperature-dependent chemical matrices. Using an integrated computational framework, we evaluated 42 such indices on a dataset of 22 lower polycyclic aromatic hydrocarbons (PAHs), targeting normal boiling point and standard enthalpy of formation. Comparative testing revealed that while several Estrada-type indices underperformed, five spectral descriptors achieved exceptional multiple correlations (MCVs[Formula: see text]). These indices were validated through regression modeling and applied to predict the properties of linear polyacenes, confirming their reliability and robustness. These findings establish temperature-based spectral invariants as powerful and generalizable tools for predictive modeling in cheminformatics.