DPT benchmark on toy GRN

Notebook benchmarks latent time inference using DPT on toy GRN data.

Library imports

from tqdm import tqdm

import numpy as np
import pandas as pd

import scanpy as sc

from rgv_tools import DATA_DIR
from rgv_tools.benchmarking import get_data_subset, get_time_correlation
from rgv_tools.core import read_as_dask

Constants

DATASET = "toy_grn"

SAVE_DATA = True
if SAVE_DATA:
    (DATA_DIR / DATASET / "results").mkdir(parents=True, exist_ok=True)

Function definitions

Data loading

adata = read_as_dask(store=DATA_DIR / DATASET / "raw" / "adata.zarr", layers=[])
adata

Pseudotime pipeline

time_correlation = []

for dataset in tqdm(adata.obs["dataset"].cat.categories):
    adata_subset = get_data_subset(adata=adata, column="dataset", group=dataset, uns_keys=[])

    adata_subset.uns["iroot"] = np.flatnonzero(adata_subset.obs["true_time"] == 0)[0]

    sc.pp.neighbors(adata_subset)
    sc.tl.diffmap(adata_subset)
    sc.tl.dpt(adata_subset)

    time_correlation.append(
        get_time_correlation(
            ground_truth=adata_subset.obs["true_time"], estimated=adata_subset.obs["dpt_pseudotime"].values
        )
    )

Data saving

if SAVE_DATA:
    pd.DataFrame({"time": time_correlation}).to_parquet(path=DATA_DIR / DATASET / "results" / "dpt_correlation.parquet")