The performance of deep generative models for learning joint embeddings of single-cell multi-omics data

Abstract

Recent extensions of single-cell studies to multiple data modalities raise new questions regarding experimental design. For example, the challenge of sparsity in single-omics data might be partly resolved by compensating for missing information across modalities. In particular, deep learning approaches, such as deep generative models (DGMs), can potentially uncover complex patterns via a joint embedding. Yet, this also raises the question of sample size requirements for identifying such patterns from single-cell multi-omics data. Here, we empirically examine the quality of DGM-based integrations for varying sample sizes. We first review the existing literature and give a short overview of deep learning methods for multi-omics integration. Next, we consider eight popular tools in more detail and examine their robustness to different cell numbers, covering two of the most common multi-omics types currently favored. Specifically, we use data featuring simultaneous gene expression measurements at the RNA level and protein abundance measurements for cell surface proteins (CITE-seq), as well as data where chromatin accessibility and RNA expression are measured in thousands of cells (10x Multiome). We examine the ability of the methods to learn joint embeddings based on biological and technical metrics. Finally, we provide recommendations for the design of multi-omics experiments and discuss potential future developments.

In situ veritas: combining omics and multiplex imaging can facilitate the detection and characterization of cell-cell interactions in tissues

“…This reconstruction of “highly probable signaling networks” is often based on scRNA-seq data, without exact pairing transcriptomic quantification with probability-based protein identification ( 30 ). Models capable of indexing both transcriptomes and epitopes by sequencing, such as CITE-seq, already exist ( 31 ) and will help to combine RNA-data and protein abundance in a contextual manner. However, as Alexander F. Schier has already asked: … “Is “landscapes” even the proper analogy for multidimensional phenotypic complexity?…”

Optimizing Sales Prediction in Online Grocery Shopping using Product Embeddings

“…For a dataset to be embeddable, it must have a column containing information that uniquely identifies that product, such as product ID [29]. The identifier ensures that the embeddings can be accurately linked with the corresponding products during training, querying, or making predictions thereby allowing for the preservation of the integrity of the embeddings and their association back to the original products in the dataset [30].…”

An in-depth comparison of linear and non-linear joint embedding methods for bulk and single-cell multi-omics

“…In another study, Brombacher et al . compared the performance of several deep-learning-based joint representation learners as a function of the number of cells in the datasets [ 27 ]. This study was focused on single-cell data and did not compare to any well-established linear methods.…”