Application of the mol2vec Technology to Large-size Data Visualization and Analysis.

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning

Background:

• Generative Topographic Mapping (GTM) is a dimensionality reduction technique valuable for data visualization and structure-activity relationship (SAR) modeling.
• High dimensionality in molecular descriptor spaces can lead to significant computational costs and slow GTM construction.

Purpose of the Study:

• To investigate alternative dimensionality reduction methods for large molecular descriptor spaces.
• To evaluate the efficacy of mol2vec as a preprocessing tool for GTM in bioactivity classification.

Main Methods:

• Dimensionality reduction of ISIDA fragment descriptors and Morgan fingerprints using Principal Component Analysis (PCA) and mol2vec.
• Training and evaluating GTM models with descriptors reduced by both PCA and mol2vec for bioactivity classification tasks.

Main Results:

• Mol2vec significantly reduced the dimensionality of descriptor spaces.
• Mol2vec-based GTM training was substantially faster compared to PCA-based GTM.
• The predictive performance of GTM in bioactivity classification was maintained when using mol2vec for dimensionality reduction.

Conclusions:

• Mol2vec offers an efficient alternative to PCA for reducing dimensionality in large molecular descriptor spaces for GTM.
• This approach accelerates GTM training without compromising its predictive accuracy for SAR studies.