dynamo.pd.tree_model(adata, group, progenitor, terminators, basis='umap', n_neighbors=30, neighbor_key=None, graph_mat=None, state_graph_method='vf', prune_graph=True, row_norm=True)[source]

This function learns a tree model of cell states (types).

It is based on the shortest path from the source to target cells of the pruned vector field based cell-type transition graph. The pruning was done by restricting cell state transition that are only between cell states that are nearby in gene expression space (often low gene expression space).

  • adata (AnnData) – AnnData object.

  • group (str) – Cell graph that will be used to build transition graph and lineage tree.

  • progenitor (str) – The source cell type name of the lineage tree.

  • terminators (List[str]) – The terminal cell type names of the lineage tree.

  • basis (str) – The basis that will be used to build the k-nearest neighbor graph when neighbor_key is not set.

  • n_neighbors (int) – The number of neighbors that will be used to build the k-nn graph, passed to dyn.tl.neighbors function. Not used when neighbor_key provided.

  • neighbor_key (Optional[str]) – The nearest neighbor graph key in adata.obsp. This nearest neighbor graph will be used to build a gene-expression space based cell-type level connectivity graph.

  • state_graph_method (str) – Method that will be used to build the initial state graph.

  • prune_graph (bool) – Whether to prune the transition graph based on cell similarities in basis bases first before learning tree model.

  • row_norm (bool) – Whether to normalize each row so that each row sum up to be 1. Note that row, columns in transition matrix correspond to source and targets in dynamo by default.


The final tree model of cell groups. See following example on how to visualize the tree via dynamo.

Return type:



>>> import dynamo as dyn
>>> adata = dyn.sample_data.pancreatic_endocrinogenesis()
>>> dyn.pp.recipe_monocle(adata)
>>> dyn.tl.dynamics(adata)
>>> dyn.tl.cell_velocities(adata)
>>> dyn.vf.VectorField(adata, basis='umap', pot_curl_div=False)
>>> dyn.pd.state_graph(adata, group='clusters', basis='umap')
>>> res = dyn.pd.tree_model(adata, group='clusters', basis='umap')
>>> # in the following we first copy the state_graph result to a new key and then replace the `group_graph` key of
>>> # the state_graph result and visualize tree model via dynamo.
>>> adata.obs['clusters2'] = adata.obs['clusters'].copy()
>>> adata.uns['clusters2_graph'] = adata.uns['clusters_graph'].copy()
>>> adata.uns['clusters2_graph']['group_graph'] = res
>>> dyn.pl.state_graph(adata, group='clusters2', keep_only_one_direction=False, transition_threshold=None,
>>> color='clusters2', basis='umap', show_legend='on data')