方法分享--空转配受体分析之LARIS

原创

追风少年i

发布于 2025-12-20 15:43:22

2950

作者，Evil Genius

比起别人真的坎坷了，干啥也不顺。

当然了，相逢既是缘分，我希望大家能学会技能，不要走我踩过的坑，偶尔罗嗦了一点，当然点到即止，科研的路还是要大家自己亲自去走。

今天我们分享一个方法--LARIS，参考文献在

其中最核心的地方在于结合空间位置信息分析配受体。

LARIS（空间转录组学中的配体与受体互作分析），这是一种精确且可扩展的方法，能够在单细胞或珠分辨率上识别细胞类型特异性及空间受限的配体-受体相互作用。LARIS兼容所有空间转录组技术，可量化互作特异性、推断信号发送-接收方向性，并检测差异相互作用如何随时间和空间变化。

LARIS通过结合计算模拟的配体-受体扩散过程与基于空间k近邻图的局部细胞异质性分析，以余弦相似度为度量，推断单细胞水平的配体-受体互作评分及空间特异性细胞间通讯。

方法看起来值得研究一下。

看看文献中的运用示例

将LARIS应用于两个真实空间数据集——人扁桃体（揭示生发中心功能分区和关键信号枢纽）与发育中小鼠大脑皮层（解析细胞互作的时空动态规律），证明了其在复杂生物系统中发现新颖生物学洞察的能力。

LARIS对人扁桃体空间数据的分析结果

大家注意依据通讯信息分析热点区域的这个分析方法。

华大数据的运用，通讯的时空变化。

这个方法主要针对的高精度，全流程的python代码。

看其中一个示例

import numpy as np
import pandas as pd
import scanpy as sc
import matplotlib.pyplot as plt
import warnings
warnings.simplefilter(action='ignore', category=FutureWarning)

# Colours
from matplotlib import cm
from matplotlib import colors, colorbar
cmap_own = cm.get_cmap('magma_r', 256)
newcolors = cmap_own(np.linspace(0,0.75 , 256))
Greys = cm.get_cmap('Greys_r', 256)
#newcolors[:1, :] = Greys(np.linspace(0.8125, 0.8725, 1))
newcolors[:10, :] = Greys(np.linspace(0.8125, 0.8725, 10))
pos_cmap = colors.ListedColormap(newcolors)

adata=sc.read('/data1/mdai/Result/single-cell/Methods/LARIS/tutorial/adata_tonsil.h5ad') 

# Check spatial mapping
x_width=adata.obsm['X_spatial'][:,0].max()-adata.obsm['X_spatial'][:,0].min()
y_width=adata.obsm['X_spatial'][:,1].max()-adata.obsm['X_spatial'][:,1].min()

plt.rcParams['figure.figsize'] = 5, 5*y_width/x_width

sc.pl.embedding(
    adata,
    basis='X_spatial',
    color=['cell_type'],
    size=80,frameon=False)

plt.rcParams['figure.figsize'] = 4, 4

# Check spatial mapping
x_width=adata.obsm['X_spatial'][:,0].max()-adata.obsm['X_spatial'][:,0].min()
y_width=adata.obsm['X_spatial'][:,1].max()-adata.obsm['X_spatial'][:,1].min()

plt.rcParams['figure.figsize'] = 5, 5*y_width/x_width

sc.pl.embedding(
    adata,
    basis='X_spatial',
    color=['cell_type'],
    groups=['T_CD4'],
    size=80,frameon=False)

plt.rcParams['figure.figsize'] = 4, 4

# Check spatial UMAP
x_width=adata.obsm['X_umap'][:,0].max()-adata.obsm['X_umap'][:,0].min()
y_width=adata.obsm['X_umap'][:,1].max()-adata.obsm['X_umap'][:,1].min()

plt.rcParams['figure.figsize'] = 5, 5*y_width/x_width

sc.pl.embedding(
    adata,
    basis='X_umap',
    color=['cell_type'],
    size=40,frameon=False)

plt.rcParams['figure.figsize'] = 4, 4

lr_df=pd.read_csv('/data1/mdai/Result/single-cell/Methods/LARIS/tutorial/human_lr_database_CellChatDB_formatted_v2.csv',index_col=0)

rows_keep=np.logical_and( lr_df['ligand'].isin(adata.var_names), lr_df['receptor'].isin(adata.var_names) )
lr_df=lr_df.loc[rows_keep].copy()

%%time
lr_adata=la.tl.prepareLRInteraction(
    adata, # object used to construt the LR object
    lr_df, # dataframe to use for interactions
    number_nearest_neighbors=20,
    use_rep_spatial='X_spatial'
)

%%time
import warnings
warnings.filterwarnings("ignore")
LARIS_variable_interactions, res_LARIS=la.tl.runLARIS(
    lr_adata,
    adata,
    use_rep='X_spatial',
    n_nearest_neighbors=20,
    random_seed=27,
    n_repeats=5,
    mu = 0.40, # larger value indicates more consideration for specificity
    sigma=100,
    remove_lowly_expressed=False,
    expressed_pct=0.1,
    n_cells_expressed_threshold=100,
    n_top_lr=lr_adata.shape[1],

    by_celltype=True,
    ### Parameters for cell type-level inference when by_celltype set to True
    groupby='cell_type', # label to group by
    use_rep_spatial='X_spatial', # spatial coordinates
    mu_celltype=100, # higher value puts more emphasis on cell type specificity
    expressed_pct_celltype=0.1, # expression percentage cut off per cell type
    remove_lowly_expressed_celltype=False,
    mask_threshold=1e-6, # mask for cosg, lower value is less restrictive
    n_neighbors_permutation=30,
    score_threshold= 1e-10,
    spatial_weight = 3.0,
)

sc.pl.dotplot(lr_adata,
              LARIS_variable_interactions.index[:10],
              groupby='region_name',
             standard_scale='var',
             cmap='Spectral_r')

x_width=adata.obsm['X_spatial'][:,0].max()-adata.obsm['X_spatial'][:,0].min()
y_width=adata.obsm['X_spatial'][:,1].max()-adata.obsm['X_spatial'][:,1].min()

plt.rcParams['figure.figsize'] = 2.5, 2.5*y_width/x_width

sc.pl.embedding(
    lr_adata,
    basis='X_spatial',
    color=LARIS_variable_interactions.index[:4],
    cmap=pos_cmap,
 #   vmin=1,
  #  vmax=1.5,
      # palette=self_palette2,
       ncols=4,
    # size=80,
    frameon=False)
plt.rcParams['figure.figsize'] = 4, 4

x_width=adata.obsm['X_spatial'][:,0].max()-adata.obsm['X_spatial'][:,0].min()
y_width=adata.obsm['X_spatial'][:,1].max()-adata.obsm['X_spatial'][:,1].min()

plt.rcParams['figure.figsize'] = 2.5, 2.5*y_width/x_width

sc.pl.embedding(
    lr_adata,
    basis='X_spatial',
    color=res_LARIS['interaction_name'][:12].unique(),
    cmap=pos_cmap,
 #   vmin=1,
  #  vmax=1.5,
      # palette=self_palette2,
       ncols=3,
      # size=80,
    frameon=False)
plt.rcParams['figure.figsize'] = 4, 4