Comparing predictive ability of QSAR/QSPR models using 2D and 3D molecular representations

Area of Science:

• Computational Chemistry
• Cheminformatics
• Drug Discovery

Background:

• Quantitative Structure-Activity Relationship (QSAR) and Quantitative Structure-Property Relationship (QSPR) models are crucial for predicting molecular activity and properties.
• Traditional QSAR/QSPR models primarily use 2D (topological) molecular representations, potentially overlooking important 3D conformational information.
• 3D molecular representations, based on inter-molecular similarity, have shown promise in virtual screening.

Purpose of the Study:

• To integrate 3D molecular representations into QSAR/QSPR modeling for regression tasks.
• To compare the performance of 3D representations against 2D representations.
• To evaluate the impact of training data set diversity on the performance of different molecular representations.

Main Methods:

• Development and application of 3D molecular representations for QSAR/QSPR model building.
• Comparative analysis of 2D and 3D representations using various regression tasks.
• Assessment of model performance across diverse training data sets.

Main Results:

• 3D molecular representations demonstrated superior performance compared to 2D representations for predicting quantum mechanics-based properties.
• For predicting small molecule activity against biological targets, no consistent performance advantage of 3D over 2D representations was observed.
• The diversity of training data sets did not consistently influence the relative performance of 2D versus 3D representations.

Conclusions:

• 3D molecular representations offer significant advantages for specific QSAR/QSPR applications, particularly in predicting physical-chemical properties.
• The choice between 2D and 3D representations may depend on the specific prediction task and data characteristics.
• Further research is warranted to fully elucidate the optimal application domains for 3D molecular representations in cheminformatics.