MultiMAP: Dimensionality Reduction and Integration of Multimodal Data

Abstract

Multimodal data is rapidly growing in many fields of science and engineering, including single-cell biology. We introduce MultiMAP, an approach for dimensionality reduction and integration of multiple datasets. MultiMAP recovers a single manifold on which all of the data resides and then projects the data into a single low-dimensional space so as to preserve the structure of the manifold. It is based on a framework of Riemannian geometry and algebraic topology, and generalizes the popular UMAP algorithm1 to the multimodal setting. MultiMAP can be used for visualization of multimodal data, and as an integration approach that enables joint analyses. MultiMAP has several advantages over existing integration strategies for single-cell data, including that MultiMAP can integrate any number of datasets, leverages features that are not present in all datasets (i.e. datasets can be of different dimensionalities), is not restricted to a linear mapping, can control the influence of each dataset on the embedding, and is extremely scalable to large datasets. We apply MultiMAP to the integration of a variety of single-cell transcriptomics, chromatin accessibility, methylation, and spatial data, and show that it outperforms current approaches in preservation of high-dimensional structure, alignment of datasets, visual separation of clusters, transfer learning, and runtime. On a newly generated single-cell Assay for Transposase-Accessible Chromatin using sequencing (scATAC-seq) and single-cell RNA-seq (scRNA-seq) dataset of the human thymus, we use MultiMAP to integrate cells along a temporal trajectory. This enables the quantitative comparison of transcription factor expression and binding site accessibility over the course of T cell differentiation, revealing patterns of transcription factor kinetics.