Fig. 1: Overview of CellNavi. | Nature Cell Biology

Fig. 1: Overview of CellNavi.

From: CellNavi predicts genes directing cellular transitions by learning a gene graph-enhanced cell state manifold

Fig. 1

a, A conceptual illustration of CellNavi’s task. Given a pair of source and target cells undergoing a transition induced by stimuli, CellNavi predicts the driver gene responsible for this transition. b, The workflow of CellNavi. The CMM maps the source and target cells onto a coordinate space of the cell manifold. The DGP then uses the cell coordinates produced by the CMM to rank the candidate genes by likelihood scores. c, An illustration of the cell manifold and its coordinate space. d, Data used for the CMM training. exps, experiments. e, Training of the CMM. The CMM consists of six GeneGraph Attention (attn.) layers designed to incorporate graph-based information. During training, single-cell transcriptomic profiles are randomly sampled from the curated HCA dataset and used as input. Cell embeddings generated by the model are then used by a transformer decoder to reconstruct gene expression profiles. f, Data used for the DGP training. g, Application scenarios and test cases of CellNavi. MoA, mechanism of action. Schematic elements created with BioRender.com.

Back to article page