Lipidome visualisation, comparison, and analysis in a vector space.

Area of Science:

• Lipidomics
• Bioinformatics
• Computational Chemistry

Background:

• Lipid structures are complex and require effective methods for classification and analysis.
• Current methods for visualizing and comparing lipidomes can be limited in scope and interpretability.

Purpose of the Study:

• To develop a computational method for embedding lipid structures into a low-dimensional space.
• To create user-friendly software for visualizing and analyzing lipidomic data.
• To validate the embedding method using established lipid databases and published datasets.

Main Methods:

• A shallow neural network was employed to generate vector embeddings for lipid structures.
• The embedding aimed to ensure that structurally similar lipids are represented by similar vectors in the chosen dimensional space.
• The web-based software, Lipidome Projector, was developed to visualize these embeddings as 2D or 3D scatterplots.

Main Results:

• The neural network successfully embedded lipid structures, with resulting distributions aligning with conventional lipid classifications.
• Lipidome Projector enables rapid exploratory analysis, quantitative comparison, and structural-level interpretation of user-provided lipidomic data.
• Qualitative comparisons with published datasets demonstrated the method's utility in interpreting complex lipidomic information.

Conclusions:

• The developed neural network embedding method provides an effective approach for organizing and visualizing lipid structures.
• Lipidome Projector serves as a valuable tool for researchers in lipidomics, facilitating data interpretation and discovery.
• This approach enhances the understanding of lipid relationships and facilitates the analysis of complex lipidomic datasets.