Failed to remove batch effect

[Somnvs]

Hello, I have trouble when I analyze my spatial transcriptome data. I follow the official course to remove batch effect(https://stereopy.readthedocs.io/en/latest/Tutorials%28Multi-sample%29/Batch_Effect.html#Integrating), however, the results of my analysis show that the samples are separated and do not overlap.

I’m wondering if there’s some specific parameter I overlooked, which is causing the results to be quite bizarre.

The script for my analysis is as follows:

import stereo as st
import warnings
warnings.filterwarnings('ignore')

import csv
import os

import pandas as pd

# outPrefix="/home/weigs/Project/01_LilinSpatial/14_mergeSurgery/10_filter9SampleBin40/result"
# outDir="/home/weigs/Project/01_LilinSpatial/14_mergeSurgery/10_filter9SampleBin40"
outPrefix="/home/weigs/Project/04_Liuzk_spatial/06_stRnaMergeSample/01_mergeSampleBin100/result"
outDir="/home/weigs/Project/04_Liuzk_spatial/06_stRnaMergeSample/01_mergeSampleBin100"
# tableFile="/home/weigs/Project/01_LilinSpatial/14_mergeSurgery/000fileListFirstColumn.txt"
binSize=100
# binSize=50

neighborNum=6 # default 6

pcNum=30 # default 30
findNeighborNum=10 # default 10

gem1="/home/weigs/Project/04_Liuzk_spatial/01_dataInfo/02_stR_data/D01567F5.tissue.gem.gz"
gem2="/home/weigs/Project/04_Liuzk_spatial/01_dataInfo/02_stR_data/D01656A1.tissue.gem.gz"
gem3="/home/weigs/Project/04_Liuzk_spatial/01_dataInfo/02_stR_data/D01656B3.tissue.gem.gz"
gem4="/home/weigs/Project/04_Liuzk_spatial/01_dataInfo/02_stR_data/D01656C3.tissue.gem.gz"


data1 = st.io.read_gem(file_path=gem1, bin_size=binSize)
data2 = st.io.read_gem(file_path=gem2, bin_size=binSize)
data3 = st.io.read_gem(file_path=gem3, bin_size=binSize)
data4 = st.io.read_gem(file_path=gem4, bin_size=binSize)


data1.tl.cal_qc()
data2.tl.cal_qc()
data3.tl.cal_qc()
data4.tl.cal_qc()


data = st.utils.data_helper.merge(data1, data2, data3, data4)
# check the shape of merged data
data.shape

# Since normalization will change the expression matrix, save raw data beforehand.
data.tl.raw_checkpoint()

data.tl.normalize_total()
data.tl.log1p()

data.tl.pca(use_highly_genes=False, n_pcs=pcNum, res_key='pca')
# data.tl.pca(use_highly_genes=True, n_pcs=pcNum, res_key='pca')
data.tl.batches_integrate(pca_res_key='pca', res_key='pca_integrated')

data.tl.neighbors(pca_res_key='pca_integrated', n_pcs=pcNum, res_key='neighbors_integrated',n_neighbors=findNeighborNum)
# compute spatial neighbors
data.tl.spatial_neighbors(
       neighbors_res_key='neighbors_integrated',
       res_key='spatial_neighbors_integrated',
       n_neighbors=neighborNum
       )

data.tl.umap(pca_res_key='pca_integrated', neighbors_res_key='spatial_neighbors_integrated', res_key='umap_integrated')
data.plt.batches_umap(res_key='umap_integrated')

[tanliwei-coder]

Try to run umap with neighbors instead of spatial_neighbors.

[Somnvs]

Try to run umap with neighbors instead of spatial_neighbors.

Thank you for your response. I had try run umap with neighbors instead of spatial_neighbors, but the results didn't change significantly.

[tanliwei-coder]

Let's see what the spatial distribution of the four samples is like, run data.plt.spatial_scatter on the data object merged from four samples:

data.plt.spatial_scatter(reorganize_coordinate=2)

[Somnvs]

plot spatial scatter, result was as follow:

[tanliwei-coder]

Run BatchQC to check whether batch effect removal is needed.

[Somnvs]

Sorry for the delayed response over the past two days; I took some time to run the BatchQC program. The final results indicate that batch effect correction is necessary for this dataset. Attached is the BatchQC report.
BatchQC_report_raw.zip

[tanliwei-coder]

Sorry for replying so late, our batch effect removal is based on harmonypy, and the paper about this algorithm is here, I guess this algorithm could not work on all data

[Somnvs]

It's ok, I hope you can develop and integrate more algorithms and technologies in the future to solve this issue that cannot be handled.