Basic tutorial

Dataset by Shen-Orr et al 2010 (Nat Methods), avaiable under GSE19830. Samples from brain, liver and lung were mixed in different proportions and analyzed using transcriptome microarray.

import anndata
import matplotlib.pyplot as plt

import dtangle

adata_sum = anndata.read_h5ad("../../examples/data/GSE19830_sum.h5ad")
adata_sum

AnnData object with n_obs × n_vars = 42 × 13934
    obs: 'sample_name', 'cel_version', 'nrows', 'ncols', 'algorithm', 'sample_geo_accession', 'sample_name_meta', 'sample_title', 'sample_source_name_ch1', 'sample_organism_ch1', 'sample_platform_id', 'sample_series_id', 'file', 'file_size', 'file_url', 'accession', 'fraction_liver', 'fraction_brain', 'fraction_lung'
    var: 'gene_id', 'n_probes', 'converged'
    uns: 'normalization', 'summarization'

Split and annotate dataset. References anndata object is generated from pure samples.

adata_ref = adata_sum[
    (adata_sum.obs["fraction_liver"] == 1)
    | (adata_sum.obs["fraction_brain"] == 1)
    | (adata_sum.obs["fraction_lung"] == 1)
]

adata_source = adata_sum[
    (adata_sum.obs["fraction_liver"] != 1)
    & (adata_sum.obs["fraction_brain"] != 1)
    & (adata_sum.obs["fraction_lung"] != 1)
]

adata_ref.obs["ref_organ"] = None
adata_ref.obs.loc[adata_ref.obs["fraction_liver"] == 1, "ref_organ"] = "liver"
adata_ref.obs.loc[adata_ref.obs["fraction_brain"] == 1, "ref_organ"] = "brain"
adata_ref.obs.loc[adata_ref.obs["fraction_lung"] == 1, "ref_organ"] = "lung"

assert adata_ref.n_obs + adata_source.n_obs == adata_sum.n_obs

/tmp/ipykernel_23758/2424560204.py:14: ImplicitModificationWarning: Trying to modify attribute `.obs` of view, initializing view as actual.
  adata_ref.obs["ref_organ"] = None

Run deconvolution

dtangle.deconvolut(
    adata_source,
    references=adata_ref,
    reference_annotation_col="ref_organ",
    data_type="microarray-gene",
    n_markers=20,
)

/home/malte/Dokumente/Github/dtangle/src/dtangle/_deconvolution.py:116: ImplicitModificationWarning: Setting element `.obsm['dtangle']` of view, initializing view as actual.
  adata.obsm[key_added] = estimates_df

Plot estimated vs. true fractions

fig, ax = plt.subplots(figsize=(6, 6))
for cell_type, color in zip(
    ["brain", "lung", "liver"],
    ["tab:blue", "tab:green", "tab:red"],
    strict=True,
):
    ax.scatter(
        x=adata_source.obs[f"fraction_{cell_type}"],
        y=adata_source.obsm["dtangle"][cell_type],
        c=color,
        label=cell_type,
    )
ax.legend(
    frameon=False,
    loc="upper left",
    bbox_to_anchor=(1, 1),
)
ax.set_xlim(0, 1)
ax.set_ylim(0, 1)

ax.plot([0, 1], [0, 1], c="gray", ls="--")

ax.set_xlabel("True fraction")
ax.set_ylabel("Estimated fraction")

Text(0, 0.5, 'Estimated fraction')