Bridging the gap: integrating advanced data science with untargeted grape metabolomics

Abstract

The unbiased characterization of grape and wine matrices through metabolomics generates vast and high-dimensional datasets which potentially represent strong foundations to expand our understanding of complex chemical and biological interactions.

The complexity and the size of these datasets challenge traditional interpretation and new tools are needed to visualize and decipher this complexity. These approaches are often developed in seemingly far domains like computer science and big data analytics, and it is important to share them with the analytical and biological community to fully profit from the information content of metabolomics investigations.

In the presentation we will focus on non-linear visualization strategies like Uniform Manifold Approximation and Projection (UMAP), showing how they can be used to reveal specific compositional patterns inside a large-scale untargeted metabolomics dataset collected for the characterization of the “grape metabolome”. In contrast to established untargeted linear visualization approaches like PCA, the UMAP representation is designed to preserve both local and global data structures offering a more “global” representation of the compositional diversity of the dataset.