Computational Methods for Single-cell Multi-omics Integration and Alignment

Abstract

Recently developed technologies to generate single-cell genomic data have made a revolutionary impact in the field of biology. Multi-omics assays offer even greater opportunities to understand cellular states and biological processes. The problem of integrating different omics data with very different dimensionality and statistical properties remains, however, quite challenging. A growing body of computational tools is being developed for this task, leveraging ideas ranging from machine translation to the theory of networks, and represents another frontier on the interface of biology and data science. Our goal in this review is to provide a comprehensive, up-to-date survey of computational techniques for the integration of single-cell multi-omics data, while making the concepts behind each algorithm approachable to a non-expert audience.

Keywords: Integration; Machine learning; Multi-omics; Single-cell; Unsupervised learning.

Figure 1

Integration and alignment of multi-omics data Multi-omics data can sometimes be sequenced from the same set of single cells (left); at other times, only the data sequenced from the same/similar sample, but different single cells are available (right). In the former case, we have the task of integrating the different data modalities (left); in the latter case, we need to first identify similar cells across the samples (right). This is the computational task of alignment.

Figure 2

Single-cell multi-omics integration methods Illustration of some common integration approaches for single-cell multi-omics data: matrix factorization uncovering a representation of both cells and omics via factors (A), neural networks which combine different -omics into a single cell representation (B), and network-based approaches, which represent cells as nodes on the graphs connected to nearby cells (C).

Figure 3

Single-cell multi-omics alignment methods Illustration of some common approaches for alignment of multi-omics single-cell data: Bayesian methods, modeling the probability distribution of -omics measurements using a number of latent variables and updating such distributions using Bayes’ formula (A), manifold alignment methods uncovering a surface in the space of omics on which the alignment can be performed (B), and neural network-based models, creating latent representations of different -omics data, which can be more easily aligned (C).

Figure 4

Distance-based alignment Schematic overview of the distance-based alignment algorithm: cells are represented by nodes in two different graph representations, corresponding to two different omics assays. Cells with very similar omics measurements are connected to form graphs. Two graphs are then aligned in order to preserve a notion of distance on the graph.