GRN formation

[1]:

from grnet.clusters import CellClasses
from grnet.models import PC, PretrainedModel
from grnet.plot import grnplot
from grnet.toydata import load_dataset, load_metadata
import matplotlib.pyplot as plt

data loading

[2]:

data = load_dataset("prototype1")
meta = load_metadata("prototype1")

lst_data = [
    data.loc[meta.cluster == v, :].iloc[:, :5] for v in meta.cluster.unique()
]

GRN formation with PC algorithm

[3]:

models = [PC(v) for v in lst_data]

for v in models:
    v.estimate()

INFO:pgmpy: Datatype (N=numerical, C=Categorical Unordered, O=Categorical Ordered) inferred from data:
 {'gene_1': 'N', 'gene_2': 'N', 'gene_3': 'N', 'gene_4': 'N', 'gene_5': 'N'}
INFO:pgmpy: Datatype (N=numerical, C=Categorical Unordered, O=Categorical Ordered) inferred from data:
 {'gene_1': 'N', 'gene_2': 'N', 'gene_3': 'N', 'gene_4': 'N', 'gene_5': 'N'}
INFO:pgmpy: Datatype (N=numerical, C=Categorical Unordered, O=Categorical Ordered) inferred from data:
 {'gene_1': 'N', 'gene_2': 'N', 'gene_3': 'N', 'gene_4': 'N', 'gene_5': 'N'}
INFO:pgmpy: Datatype (N=numerical, C=Categorical Unordered, O=Categorical Ordered) inferred from data:
 {'gene_1': 'N', 'gene_2': 'N', 'gene_3': 'N', 'gene_4': 'N', 'gene_5': 'N'}
INFO:pgmpy: Datatype (N=numerical, C=Categorical Unordered, O=Categorical Ordered) inferred from data:
 {'gene_1': 'N', 'gene_2': 'N', 'gene_3': 'N', 'gene_4': 'N', 'gene_5': 'N'}

edges are resturned as a list

[4]:

models[0].estimate()
models[0].edges

INFO:pgmpy: Datatype (N=numerical, C=Categorical Unordered, O=Categorical Ordered) inferred from data:
 {'gene_1': 'N', 'gene_2': 'N', 'gene_3': 'N', 'gene_4': 'N', 'gene_5': 'N'}

[4]:

[('gene_4', 'gene_2')]

edge information can be represented as a GRN matrix

[5]:

models[0].get_matrix()

[5]:

	gene_1	gene_2	gene_3	gene_4	gene_5
gene_1	1.0	0.0	0.0	0.0	0.0
gene_2	0.0	1.0	0.0	1.0	0.0
gene_3	0.0	0.0	1.0	0.0	0.0
gene_4	0.0	0.0	0.0	1.0	0.0
gene_5	0.0	0.0	0.0	0.0	1.0

Note: GRN matrices should fulfill the requirements as follows; 1. DxD matrix (where D: the number of genes, i.e., the number of columns) 2. rows names and columns names coincide each other 3. all elements are either 0 or 1 4. all diagonal elements are 1 5. if the GRN is an undirected graph, GRN matrix can be either upper/lower triangular

Plotting GRN

[6]:

fig, ax = plt.subplots(figsize=(4, 4))

grnplot(models[0], ax=ax)

ax.set(title="cluster_1 (prototype1)");

../_images/notebooks_pc_12_0.png

Generate `Estimator` Class from GRN matrices

use PretrainedModel class

[7]:

grn_mat = models[0].get_matrix()

pre = PretrainedModel(grn_mat)

you can access to edges/matrix

[8]:

pre.edges

[8]:

[('gene_2', 'gene_4')]

[9]:

pre.get_matrix()

[9]:

	gene_1	gene_2	gene_3	gene_4	gene_5
gene_1	1.0	0.0	0.0	0.0	0.0
gene_2	0.0	1.0	0.0	1.0	0.0
gene_3	0.0	0.0	1.0	0.0	0.0
gene_4	0.0	0.0	0.0	1.0	0.0
gene_5	0.0	0.0	0.0	0.0	1.0

for self.data, the GRN matrix will be returned

[10]:

pre.data == pre.get_matrix()

[10]:

	gene_1	gene_2	gene_3	gene_4	gene_5
gene_1	True	True	True	True	True
gene_2	True	True	True	True	True
gene_3	True	True	True	True	True
gene_4	True	True	True	True	True
gene_5	True	True	True	True	True

PretrainedModel is also visualized by grnplot function

[11]:

fig, ax = plt.subplots(figsize=(4, 4))

grnplot(pre, ax=ax, color="C1")

ax.set(title="cluster_1 (prototype1)");

../_images/notebooks_pc_22_0.png

Handle all clusters in the dataset with CellClasses

estimate GRNs for all models before assign to CellClasses

[12]:

dataset = CellClasses(
    models=models,
    names=[f"cluster_{i + 1}" for i in range(len(lst_data))],
    colors="rainbow"
)

you can visualize them by assigning cluster names or cluster ids

[13]:

fig, ax = plt.subplots(1, 5, figsize=(20, 4))

for i in range(5):
    grnplot(dataset, ax=ax[i], id=f"cluster_{i + 1}")
    ax[i].set(title=dataset.names[i])

../_images/notebooks_pc_27_0.png