Tutorial 3, Integration of simulated data

Tutorial 3, Integration of simulated data#

import sys
sys.path.append(r"/home/wangheqi/PycharmProject/")

import spcoral

from typing import Optional
import pandas as pd
import numpy as np
import scanpy as sc
import sklearn
import anndata
import torch
import matplotlib.pyplot as plt

Read the data#

The generation of the simulation data employed the method described in the article by Townes et al..

The simulated data are divided into two types: those with the same resolution pair (adata_omics1 & adata_omics2) and those with different resolution pair (adata_omics1 & adata_omics2_diff).

User can download this data from google drive.

adata_omics1 = sc.read_h5ad('/home/wangheqi/PycharmProject/data/integtarion_sim/simdata_1/adata_sim_omics1_shuffled.h5ad')
adata_omics2 = sc.read_h5ad('/home/wangheqi/PycharmProject/data/integtarion_sim/simdata_1/adata_sim_omics2_shuffled.h5ad')
adata_omics2_diff = sc.read_h5ad('/home/wangheqi/PycharmProject/data/integtarion_sim/simdata_1/adata_sim_down_omics2_shuffled_annotation.h5ad')

Show domains of the raw data#

%matplotlib inline

fig, ax = plt.subplots(figsize=(6, 6))
sc.pl.embedding(adata_omics1, basis='spatial', color='louvain', ax=ax)
../_images/4e770c8f9af972401a08740da155765e75fa4d8a7a8771ca7722443f70cb0527.png 
fig, ax = plt.subplots(figsize=(6, 6))
sc.pl.embedding(adata_omics2, basis='spatial', color='louvain', ax=ax)
../_images/91bad6ebbb6cfcf6036f43b8bf98e8c80e66914bcac9536a426e086a4425a358.png 
fig, ax = plt.subplots(figsize=(6, 6))
sc.pl.embedding(adata_omics2_diff, basis='spatial', color='louvain', ax=ax)
../_images/3bd4003963212ffb8d937be2fff84cdb85481aea6bf2133746546ac695ee0113.png

Preprocess multi-omics data#

sc.pp.scale(adata_omics1, max_value=5)
sc.pp.pca(adata_omics1, n_comps=50)
sc.pp.neighbors(adata_omics1)

sc.pp.scale(adata_omics2, max_value=5)
sc.pp.pca(adata_omics2, n_comps=30)
sc.pp.neighbors(adata_omics2, n_neighbors=30)

sc.pp.scale(adata_omics2_diff, max_value=5)
sc.pp.pca(adata_omics2_diff, n_comps=30)
sc.pp.neighbors(adata_omics2_diff, n_neighbors=50)

Model training of the same resolution#

adata_omics1.obsm['feat'] = adata_omics1.obsm['X_pca']
adata_omics2.obsm['feat'] = adata_omics2.obsm['X_pca']
adata_omics2_diff.obsm['feat'] = adata_omics2_diff.obsm['X_pca']

Model = spcoral.model.integrate_model(
    adata_omics1,
    adata_omics2,
    graph_method_single='radius',
    radius_spatial_omics1=0.2,
    radius_spatial_omics2=0.2,
    use_obsm='spatial',
    device=torch.device('cuda:0'),
    random_seed=2020
)

[Fast Mode] Seed=2020, cudnn.benchmark=True, multi-thread ON

adata_omics1, adata_omics2, loss_list = Model.train()
adata_omics1, adata_omics2 = spcoral.analysis.cluster(adata_omics1, adata_omics2, cluster_method='kmeans', cluster_number=5)

[Fast Mode] Seed=2020, cudnn.benchmark=True, multi-thread ON

Visualization of the same resolution#

sc.pl.embedding(adata_omics1, basis='spatial', color='domain', title='Omics 1')
../_images/1d47086f9d9fd36b8f48167e23f247c7604aee0ec6fbb259d9a1f8fd2b62a869.png 
sc.pl.embedding(adata_omics2, basis='spatial', color='domain', title='Omics 2')
../_images/72cfafeb49cca706d16c6bc27ccdf9367143448092151e12b38979bd1143c6a6.png

Model training of the different resolutions#

Model = spcoral.model.integrate_model(
    adata_omics1,
    adata_omics2_diff,
    graph_method_single='radius',
    radius_spatial_omics1=0.2,
    radius_spatial_omics2=0.2,
    use_obsm='spatial',
    device=torch.device('cuda:1'),
    random_seed=2020
)

[Fast Mode] Seed=2020, cudnn.benchmark=True, multi-thread ON

adata_omics1, adata_omics2_diff, loss_list = Model.train()
adata_omics1, adata_omics2_diff = spcoral.analysis.cluster(adata_omics1, adata_omics2_diff, cluster_method='kmeans', cluster_number=5)

[Fast Mode] Seed=2020, cudnn.benchmark=True, multi-thread ON

Visualization of the different resolutions#

sc.pl.embedding(adata_omics1, basis='spatial', color='domain', title='Omics 1')
../_images/8175ffaf59842638488f10b9e227378b2beaea8196b8bae111352079cece8dfd.png 
sc.pl.embedding(adata_omics2_diff, basis='spatial', color='domain', title='Omics 2 diffenet resolution')
../_images/972ccd20fa22f0d40d8193587fea87fbf1766cc5cc9bef85ffbe10ebd3c111ba.png
Contents