Variational autoencoders learn transferrable representations of metabolomics data

Abstract

Dimensionality reduction approaches are commonly used for the deconvolution of high-dimensional metabolomics datasets into underlying core metabolic processes. However, current state-of-the-art methods are widely incapable of detecting nonlinearities in metabolomics data. Variational Autoencoders (VAEs) are a deep learning method designed to learn nonlinear latent representations which generalize to unseen data. Here, we trained a VAE on a large-scale metabolomics population cohort of human blood samples consisting of over 4500 individuals. We analyzed the pathway composition of the latent space using a global feature importance score, which demonstrated that latent dimensions represent distinct cellular processes. To demonstrate model generalizability, we generated latent representations of unseen metabolomics datasets on type 2 diabetes, acute myeloid leukemia, and schizophrenia and found significant correlations with clinical patient groups. Notably, the VAE representations showed stronger effects than latent dimensions derived by linear and non-linear principal component analysis. Taken together, we demonstrate that the VAE is a powerful method that learns biologically meaningful, nonlinear, and transferrable latent representations of metabolomics data.

Statistical methods and resources for biomarker discovery using metabolomics

“…Variational autoencoders is an unsupervised deep learning method that operates by encoding input data into a non-linear, lower-dimensional latent space that can be used to reproduce the original data without loss of information. It has recently been advocated for use with metabolomics data to learn its transferable latent representations; which can help expose clusters of samples with specific metabolite levels [ 97 ].…”

Deciphering the Lexicon of Protein Targets: A Review on Multifaceted Drug Discovery in the Era of Artificial Intelligence

“…In a similar fashion, a KG from CoV-KGE contains 15 million edges from 39 distinct relation categories . Subsequently, unsupervised dimensionality reduction methods, such as multimodal autoencoder (AE), variational autoencoder (VAE), and GNN, were implemented to extract characteristics of drugs, diseases, and their associations from the KG. , These characteristics of the heterogeneous KG were utilized to prepare the data set for training, validation, and testing purposes, as well as ML and DL model construction. Currently, DL strategies are most widely used to process graph-structured data because of their capacity to manage complex network data. ,, Graph convolutional neural network was utilized by AI-DrugNet to identify drug–target associations .…”

Representation Learning to Effectively Integrate and Interpret Omics Data

“…Data integration and dimensionality reduction are central aspects of many data analysis pipelines both in bioinformatics and other fields [4]. Among many models, Variational Autoencoders (VAEs) have been shown to be useful at solving a variety of problems such as data compression, de-noising, learning transferable representations of data and data fusion [5,6,7]. Nevertheless, a great limitation of using VAEs in biology has been the inability to obtain regularised, and hence interpretable, embeddings where points close in the original feature space are also close in the latent space, limiting their use in clinical practice [8,9].…”