dpi-sc 1.1.5

An end-to-end single-cell multimodal analysis model with deep parameter inference.

Single-cell multimodal modeling with deep parametric inference

The proliferation of single-cell multimodal sequencing technologies has enabled us to understand cellular heterogeneity with multiple views, providing novel and actionable biological insights into the disease-driving mechanisms. Here, we propose the deep parametric inference (DPI) model, an end-to-end framework for single-cell multimodal data analysis. At the heart of DPI is the multimodal parameter space, where the parameters from each modal are inferred by neural networks.

The dpi framework works with scanpy and supports the following single-cell multimodal analyses

• Multimodal data integration
• Multimodal data noise reduction
• Cell clustering and visualization
• Reference and query cell types
• Cell state vector field visualization

Pip install

pip install dpi-sc

Datasets

The dataset participating in "Single-cell multimodal modeling with deep parametric inference" can be downloaded at DPI data warehouse

Tutorial

We use pbmc1k data set to demonstrate the process of DPI analysis of single cell multimodal data.

Import dependencies

import scanpy as sc
import dpi

Retina image output (optional)

%matplotlib inline
%config InlineBackend.figure_format = 'retina'

Load dataset

# The dataset can be downloaded from [Datasets] above.
sc_data = sc.read_h5ad("PBMC_COVID19_Healthy_Annotated.h5ad")

Set marker collection

rna_markers = ["CCR7", "CD19", "CD3E", "CD4"]
protein_markers = ["AB_CCR7", "AB_CD19", "AB_CD3", "AB_CD4"]

Preprocessing

dpi.preprocessing(sc_data)
dpi.normalize(sc_data, protein_expression_obsm_key="protein_expression")
sc_data.var_names_make_unique()
sc.pp.highly_variable_genes(
    sc_data,
    n_top_genes=3000,
    flavor="seurat_v3",
    subset=False
)
dpi.add_genes(sc_data, rna_markers)
sc_data = sc_data[:,sc_data.var["highly_variable"]]
dpi.scale(sc_data)

Prepare and run DPI model

Configure DPI model parameters

dpi.build_mix_model(sc_data, net_dim_rna_list=[512, 128], net_dim_pro_list=[128], net_dim_rna_mean=128, net_dim_pro_mean=128, net_dim_mix=128, lr=0.0001)

Run DPI model

dpi.fit(sc_data)

Visualize the loss

dpi.loss_plot(sc_data)

Save DPI model (optional)

dpi.saveobj2file(sc_data, "COVID19PBMC_healthy.dpi")
#sc_data = dpi.loadobj("COVID19PBMC_healthy.dpi")

Visualize the latent space

Extract latent spaces

dpi.get_spaces(sc_data)

Visualize the spaces

dpi.space_plot(sc_data, "mm_parameter_space", color="green", kde=True, bins=30)
dpi.space_plot(sc_data, "rna_latent_space", color="orange", kde=True, bins=30)
dpi.space_plot(sc_data, "pro_latent_space", color="blue", kde=True, bins=30)

Preparation for downstream analysis

Extract features

dpi.get_features(sc_data)

Get denoised datas

dpi.get_denoised_rna(sc_data)
dpi.get_denoised_pro(sc_data)

Cell clustering and visualization

Cell clustering

sc.pp.neighbors(sc_data, use_rep="mix_features")
dpi.umap_run(sc_data, min_dist=0.4)
sc.tl.leiden(sc_data)

Cell cluster visualization

sc.pl.umap(sc_data, color="leiden")

Observe multimodal data markers

RNA markers

dpi.umap_plot(sc_data, featuretype="rna", color=rna_markers, ncols=2)
dpi.umap_plot(sc_data, featuretype="rna", color=rna_markers, ncols=2, layer="rna_denoised")

Protein markers

dpi.umap_plot(sc_data, featuretype="protein", color=protein_markers, ncols=2)
dpi.umap_plot(sc_data, featuretype="protein", color=protein_markers, ncols=2, layer="pro_denoised")

Reference and query

Reference objects need to be pre-set with cell labels.

sc.pl.umap(sc_data, color="initial_clustering", frameon=False, title="PBMC COVID19 Healthy labels")

# The dataset can be downloaded from [Datasets] above.
filepath = "/home/hh/bigdata/hh/DPI/COVID-19/COVID19_Asymptomatic.h5ad"
sc_data_COVID19_Asymptomatic = sc.read_h5ad(filepath)

dpi.normalize(sc_data_COVID19_Asymptomatic, protein_expression_obsm_key="protein_expression")

sc_data_COVID19_Asymptomatic = sc_data_COVID19_Asymptomatic[:,sc_data.var.index]

sc_data_COVID19_Asymptomatic.obsm["rna_nor"] = sc_data.mm_rna.transform(sc_data_COVID19_Asymptomatic.X).astype("float16")
sc_data_COVID19_Asymptomatic.obsm["pro_nor"] = sc_data.mm_pro.transform(sc_data_COVID19_Asymptomatic.obsm["pro_nor"]).astype("float16")

dpi.annotate(sc_data, ref_labelname="initial_clustering", sc_data_COVID19_Asymptomatic)

sc.pl.umap(sc_data_COVID19_Asymptomatic, color="labels", frameon=False, title="PBMC COVID19 Asymptomatic Annotated")

sc.pl.umap(sc_data_COVID19_Asymptomatic, color="initial_clustering", frameon=False, title="PBMC COVID19 Asymptomatic labels")