Fig. 2

The graphical model of SHARE-Topic: graphical representation of the SHARE-Topic model illustrating the interrelationships between latent topics and observed gene expression reads (\(n_g^c\)) and chromatin regions observed (\(r^c\)). The model depicts the interactions on a given cell c between its transcriptomic profile and accessible chromatin region profile. These observations, according to SHARE-Topic are generated in the following way: each cell c is a different mixture of topics (\(\theta ^c_t\)). Given a contribution of a certain topic t, there is a likelihood to observe a gene count in the cell \(n_g^c\) sampled from a Poisson distribution with an expected number of reads \(\lambda _g^t\). On the other side also for a given topic t contribution in a cell, the likelihood of finding a region \(r^c\) open is \(\phi _r^t\). The priors are shown in the model at the top layer and descend down in a hierarchical fashion to observations