Tutorial 1: Data and Inputs¶

scTenifoldpy expects expression data as genes x cells. The public API accepts either a pandas.DataFrame in that shape or an AnnData-like object with X, var_names, and obs_names.

In [1]:

import pandas as pd
from scTenifold.data import get_test_df

x = get_test_df(n_cells=80, n_genes=120, random_state=0)
y = get_test_df(n_cells=80, n_genes=120, random_state=1)

x.shape, y.shape

import pandas as pd from scTenifold.data import get_test_df x = get_test_df(n_cells=80, n_genes=120, random_state=0) y = get_test_df(n_cells=80, n_genes=120, random_state=1) x.shape, y.shape

Out[1]:

((120, 80), (120, 80))

Rows are genes and columns are cells. Gene names are required because downstream steps align networks by shared gene names.

In [2]:

x.index[:5].tolist(), x.columns[:5].tolist()

x.index[:5].tolist(), x.columns[:5].tolist()

Out[2]:

(['MT-1', 'MT-2', 'MT-3', 'MT-4', 'MT-5'],
 ['Cell-1', 'Cell-2', 'Cell-3', 'Cell-4', 'Cell-5'])

AnnData inputs¶

AnnData stores observations as rows and variables as columns, so scTenifoldpy transposes adata.X internally to genes x cells. Use layer="counts" in the high-level API to read an AnnData layer instead of adata.X.

In [3]:

from scTenifold.core._networks import anndata_to_dataframe

class MiniAnnData:
    def __init__(self, matrix, obs_names, var_names):
        self.X = matrix
        self.obs_names = obs_names
        self.var_names = var_names
        self.layers = {}

adata = MiniAnnData(x.T.to_numpy(), obs_names=x.columns, var_names=x.index)
converted = anndata_to_dataframe(adata)
converted.equals(x)

from scTenifold.core._networks import anndata_to_dataframe class MiniAnnData: def __init__(self, matrix, obs_names, var_names): self.X = matrix self.obs_names = obs_names self.var_names = var_names self.layers = {} adata = MiniAnnData(x.T.to_numpy(), obs_names=x.columns, var_names=x.index) converted = anndata_to_dataframe(adata) converted.equals(x)

Out[3]:

True

Remote example datasets¶

fetch_data() downloads example datasets from the companion data repository. It is useful for demos, but notebook tutorials keep remote downloads optional so the docs can build offline.

In [4]:

from scTenifold.data import list_data

# Requires network access:
# list_data()
# datasets = fetch_data("AD")

from scTenifold.data import list_data # Requires network access: # list_data() # datasets = fetch_data("AD")