Generative Topographic Mapping (GTM): Universal Tool for Data Visualization, Structure-Activity Modeling and Dataset...

Area of Science:

• Computational chemistry
• Cheminformatics
• Data science

Background:

• Dimensionality reduction techniques are crucial for analyzing complex chemical datasets.
• Existing methods like PCA and SOM have limitations in representing data distributions.
• Generative Topographic Maps (GTM) offer a novel approach to data visualization and analysis.

Purpose of the Study:

• To evaluate the utility of Generative Topographic Maps (GTM) for data visualization, structure-activity modeling, and chemical database comparison.
• To highlight the advantages of GTMs over other dimensionality reduction techniques.
• To demonstrate the application of GTMs in building classification models and comparing chemical libraries.

Main Methods:

• Utilized Generative Topographic Maps (GTM) for dimensionality reduction and data visualization.
• Employed molecular descriptors from subsets of the Database of Useful Decoys (DUD).
• Leveraged probability distribution functions (PDFs) generated by GTMs for classification and comparison.

Main Results:

• GTMs successfully provided probability distribution functions (PDFs) in both high-dimensional and 2D latent spaces.
• PDFs were effectively used to build classification models using a Bayesian approach.
• The Bhattacharyya kernel, applied to GTM-generated PDFs, provided an elegant method for global chemical library comparison.

Conclusions:

• Generative Topographic Maps (GTM) are highly effective for data visualization, structure-activity modeling, and database comparison.
• The ability of GTMs to provide probability distribution functions is a key advantage over other methods.
• GTMs combined with the Bhattacharyya kernel offer a powerful framework for analyzing and comparing chemical datasets.