NeuroMDAVIS: Visualization of single-cell multi-omics data under deep learning framework

Single-cell technologies have favoured extensive advancements in cell-type discovery, cell state identi-fication, development of lineage tracing and disease understanding among others. Further, single-cell multi-omics data generated using modern technologies provide several views of omics contribution for the same set of cells. However, dimension reduction and visualization of biological datasets (single or multi-omics) remain a challenging task since obtaining a low-dimensional embedding that preserves information about local and global structures in data, is difficult. Further, combining different views obtained from each omics layer to interpret the underlying biology is even more challenging. Earlier, we have developed NeuroDAVIS which can perform the task of visualization of high-dimensional datasets of a single modality while preserving cluster-structures within the data. Nevertheless, there is no model so far that supports joint visualization of multi-omics datasets. Joint visualization refers to transforming the feature space of each individual modality and combining them to produce a latent embedding that supports visualization of the multi-modal dataset in the newly transformed feature space. In this work, we introduce NeuroMDAVIS which is a generalized version of NeuroDAVIS for visualization of biological datasets having multiple modalities. To the best of our knowledge, NeuroMDAVIS is the first of its kind multi-modal data visualization model. It is able to learn both local and global relationships in the data while generating a low-dimensional embedding useful for downstream tasks. NeuroMDAVIS competes against state-of-the-art visualization models like t-Distributed Stochastic Neighbor Embedding (t-SNE), Uniform Manifold Approximation and Projection (UMAP), Fast interpolation-based t-SNE (Fit-SNE), and the Siamese network-based visualization method (IVIS).