preprocessing/6_baseline_sciplex_dataset.py

# ---
# jupyter:
#   jupytext:
#     text_representation:
#       extension: .py
#       format_name: light
#       format_version: '1.5'
#       jupytext_version: 1.16.1
#   kernelspec:
#     display_name: Python 3.7.12 ('chemical_CPA')
#     language: python
#     name: python3
# ---

# # 6 BASELINE SCIPLEX DATASET

# **Requires**
# sciplex_complete_middle_subset_lincs_genes.h5ad
#
# **Outputs**
# adata_baseline_high_dose.h5ad
#
#

# +
import pandas as pd
import scanpy as sc

from chemCPA.paths import DATA_DIR

pd.set_option('display.max_columns', 200)
# -

list(DATA_DIR.iterdir())

adata_sciplex = sc.read(DATA_DIR/ "sciplex_complete_middle_subset_lincs_genes.h5ad")

adata_sciplex.obs.columns

adata_sciplex.obs.loc[adata_sciplex.obs.split_ood_multi_task == 'ood', 'condition'].unique()

# +
# Subset to second largest dose

print(adata_sciplex.obs.dose.unique())
adata_sciplex = adata_sciplex[adata_sciplex.obs.dose.isin([0., 1e4])].copy()

# +
# Add new splits for dose=1000 and cell_type (A549, MCF7, K562) being unseen for ood drugs 

for cell_type in adata_sciplex.obs.cell_type.unique():
    print(cell_type)
    adata_sciplex.obs[f'split_baseline_{cell_type}'] = adata_sciplex.obs['split_ood_multi_task']
    sub_df = adata_sciplex.obs.loc[(adata_sciplex.obs[f'split_baseline_{cell_type}'] == 'ood') * (adata_sciplex.obs.cell_type != cell_type)]

    train_test = sub_df.index
    test = sub_df.sample(frac=0.5).index 

    adata_sciplex.obs.loc[train_test,f'split_baseline_{cell_type}'] = 'train'
    adata_sciplex.obs.loc[test,f'split_baseline_{cell_type}'] = 'test'
# -

adata_sciplex.obs['split_baseline_A549'].value_counts()

pd.crosstab(adata_sciplex.obs['split_ood_multi_task'], adata_sciplex.obs['condition'])

# +
# Quick check that everything is correct

cell_type = 'K562'

# pd.crosstab(adata_sciplex.obs[f'split_baseline_{cell_type}'], adata_sciplex.obs['condition'])
pd.crosstab(adata_sciplex.obs[f'split_baseline_{cell_type}'], adata_sciplex.obs['cell_type'])

# +
# write adata 

adata_sciplex.write(DATA_DIR/'adata_baseline_high_dose.h5ad', compression="gzip")
# -