QSPR modeling: graph connectivity indices versus line graph connectivity indices

Area of Science:

• * Quantitative Structure-Property Relationship (QSPR) modeling
• * Cheminformatics
• * Physical Organic Chemistry

Background:

• * QSPR models are crucial for predicting molecular properties.
• * Connectivity indices are widely used structural descriptors.
• * Previous QSPR models for alkanes utilized standard exponent values.

Purpose of the Study:

• * To reinvestigate and enhance existing QSPR models for alkanes.
• * To compare the predictive power of vertex- and edge-connectivity indices.
• * To evaluate the impact of optimized exponents and line graphs on model performance.

Main Methods:

• * Calculation of vertex- and edge-connectivity indices for alkane trees and line graphs.
• * Optimization of exponent values for connectivity indices.
• * Generation and comparison of six QSPR models per property.
• * Statistical evaluation of model performance.

Main Results:

• * QSPR models with optimized exponents significantly outperformed those with standard exponents.
• * Edge-connectivity indices showed superior performance for molecular volume-dependent properties (molar volumes, molar refractions).
• * Vertex-connectivity indices demonstrated better results for molecular surface-dependent properties (boiling points, gas chromatographic retention indices).
• * The use of line graphs did not generally improve predictive power.

Conclusions:

• * Optimized exponents enhance the accuracy of QSPR models for alkanes.
• * The choice of connectivity index (vertex vs. edge) depends on the property type.
• * Edge-connectivity indices are more suitable for volume-dependent properties.
• * Vertex-connectivity indices are more appropriate for surface-dependent properties.