MASSA Algorithm: an automated rational sampling of training and test subsets for QSAR modeling.

Area of Science:

• Computational Chemistry
• cheminformatics
• Drug Discovery

Background:

• Quantitative Structure-Activity Relationship (QSAR) models are crucial for identifying bioactive molecules.
• Effective dataset preparation, including rational sampling into training and test sets, significantly impacts QSAR model quality.
• Current methods for dataset sampling can be suboptimal, especially when descriptor availability is limited.

Purpose of the Study:

• To present MASSA, an automated Python tool for rational dataset sampling in QSAR/QSPR modeling.
• To demonstrate MASSA's ability to explore molecular spaces for improved dataset division.
• To provide a method for constructing QSAR models with reduced variability and enhanced validation metrics.

Main Methods:

• Utilizing Principal Component Analysis (PCA), Hierarchical Cluster Analysis (HCA), and K-modes for exploring molecular spaces.
• Implementing an automated algorithm for dividing datasets into training and test sets based on molecular properties.
• Generating graphical representations for data insights.

Main Results:

• MASSA enables automatic, rational dataset sampling, outperforming random methods.
• The tool facilitates the construction of multiple QSAR models using consistent training/test sets, leading to lower variability.
• Improved validation metrics were observed even when QSAR descriptors differed from those used for dataset separation.
• MASSA's applicability extends across different QSAR/QSPR techniques.

Conclusions:

• MASSA is a valuable tool for enhancing the reliability and performance of QSAR/QSPR models.
• The rational sampling approach improves model consistency and predictive accuracy.
• MASSA offers flexibility and provides valuable data visualization for cheminformatics applications.