Comparison of 2D fingerprint types and hierarchy level selection methods for structural grouping using Ward's...

Area of Science:

• * Cheminformatics
• * Computational Chemistry
• * Data Science

Background:

• * Chemical data analysis relies on effective methods for identifying structural similarities and grouping compounds.
• * Two-dimensional (2D) fingerprints are crucial for representing molecular structures in computational analyses.
• * Hierarchical clustering algorithms are widely used for grouping similar chemical compounds.

Purpose of the Study:

• * To evaluate the performance of four 2D fingerprint types (MACCS, Unity, BCI, Daylight) in representing chemical series.
• * To assess nine distinct methods for selecting optimal cluster levels from hierarchical clustering outputs.
• * To provide recommendations for selecting appropriate fingerprint types and clustering parameters for chemical data analysis.

Main Methods:

• * Utilized Ward's clustering algorithm for hierarchical clustering.
• * Applied the algorithm to subsets of the National Cancer Institute HIV dataset and a corporate compound dataset.
• * Compared the effectiveness of MACCS, Unity, BCI, and Daylight fingerprints in cluster formation.

Main Results:

• * Demonstrated variations in the ability of different fingerprint types to capture chemical series.
• * Showcased the impact of various cluster level selection methods on the interpretability of results.
• * Identified specific fingerprint and cluster level combinations that perform well for certain data types.

Conclusions:

• * The choice of 2D fingerprint significantly influences the identification of chemical series through clustering.
• * Optimal cluster level selection methods are critical for deriving meaningful insights from chemical data.
• * Findings guide the selection of robust cheminformatics and data mining strategies for chemical compound datasets.