Cobolt

Cobolt is a method developed for modeling multi-omic single-cell sequencing datasets. It provides an integrated analysis of multi-omic data by learning a latent representation shared by the different data modalities. Cobolt jointly models these multiple modalities via a novel application of Multimodal Variational Autoencoder to a hierarchical generative model. It uses a probablity model that is robust to sparse counts and high numbers of features, therefore works for omics such as chromatin accessibility or methylation status.