A comparison of different electrostatic potentials on prediction accuracy in CoMFA and CoMSIA studies

Area of Science:

• Computational chemistry
• Drug discovery and development
• Structural biology

Background:

• Accurate atomic charge assignment is critical for Quantitative Structure-Activity Relationship (QSAR) studies.
• Several methods exist for assigning molecular electrostatic potentials, but their impact on QSAR quality lacks systematic comparison.
• This study addresses this gap by evaluating various charge assignment methods in QSAR modeling.

Purpose of the Study:

• To systematically compare the performance of nine semi-empirical and empirical charge assignment methods.
• To evaluate the impact of different charge models on the predictive accuracy of Comparative Molecular Field Analysis (CoMFA) and Comparative Molecular Similarity Indices Analysis (CoMSIA) models.
• To guide the selection of optimal electrostatic potential models for QSAR studies.

Main Methods:

• Evaluated nine charge assignment methods: AM1, AM1-BCC, CFF, Formal, Gasteiger, Gasteiger-Hückel, Hückel, MMFF, and VC2003.
• Assessed method performance using standard datasets within CoMFA and CoMSIA modeling frameworks.
• Utilized prediction accuracy and cross-validation correlation coefficient (q(2)) as key evaluation criteria.

Main Results:

• The Gasteiger-Hückel method, despite common usage, showed poor prediction accuracy.
• AM1-BCC generally outperformed other methods in prediction accuracy but did not consistently yield higher q(2) values.
• The CFF charge model demonstrated the best prediction accuracy when q(2) was the primary evaluation criterion.

Conclusions:

• The choice of charge assignment method significantly influences QSAR model performance.
• The CFF charge model is recommended for CoMFA and CoMSIA studies requiring high prediction accuracy.
• This comparative analysis provides valuable insights for researchers selecting electrostatic potential models in drug design and discovery.