Source codes and a demo of JSNMF are provided in this repository.
Running environment：MATLAB R2019b or later.
The external functions used in the manuscript and JSNMF can be found in the folder external/.
jsnmf.m: jsnmf implementation, a joint single-cell multi-omics integration approach based on semi-orthogonal nonnegative matrix factorization to dissect cellular heterogeneity
preprocessing.R: data preprocessing script before running JSNMF.
parameter_selection.m: parameters selection for JSNMF(i.e. two hyperparameters alpha and gamma).
SNF.m: similarity network fusion script provided in the original publication "Similarity network fusion for aggregating data types on a genomic scale".
Wtrim.m: conduct KNN on cell-cell similairity matrix obtained from JSNMF.
RAGI.m: evaluate the performance of the clustering methods using the Residual Average Gini Index (RAGI).
plot_fig4.m: plot figure 4 in the body text.
sameH_jsnmf.m: implement JSNMF (same H) on mouse kidney data.
jsnmf_batch_correct_pari5.m: an extension of JSNMF to correct batch effect on all five Paired-Tag datasets. In this setting, there are six modalities in total and the modality of RNA is shared in the five single-cell experiments.
jsnmf_3mod.m: an extension of JSNMF to analyze single-cell multi-omics data with more than two modalities measured from the same single cell. A scNMT-seq dataset are used as an illustrative example.
parameter_selection_batcheffect_pair5.m, parameter_select_for_same_H.m and pa_sel.m: parameter selection for jsnmf_batch_correct_pari5.m, sameH_jsnmf.m and jsnmf_3mod.m, respectively.
The datasets used in this manuscript have been uploaded.
Please refer to two examples for a quick guide to JSNMF:
jsnmf_mouse brain_H3K4me3_paired-Tag.mlx: an example to use jsnmf method on mouse brain data (scRNA + H3K4me3, Paired-Tag). It contains some downstream analysis tasks, such as clustering, visualization and so on.
jsnmf_mouse kidney_sci-CAR.mlx: an example to use jsnmf method on mouse kidney data (sci-CAR). It contains some downstream analysis tasks, such as clustering, visualization and so on.
We also include a tutorial of JSNMF (same H) in jsnmf_sameH.mlx.
JSNMF is generalizable to integrate multiple single-cell multi-omics experiments and single-cell experiments with more than two molecular modalities profiled in the same cell. In the first example, we extended JSNMF to integrate all five Paired-Tag datasets (RNA + H3K27me3, RNA + H3K4me3, RNA + H3K9me3, RNA + H3K27ac and RNA + H3K4me1) simultaneously. In the second example, we extended JSNMF to analyze single-cell multi-omics data with more than two modalities measured from the same single cell. We analyzed a scNMT-seq dataset generated from the developing mouse embryo, where gene expression, chromatin accessibility and methylation were profiled simultaneously in the same cell.
The tutorials for JSNMF to integrate multiple single-cell multi-omics experiments and more molecular modalities are presented in jsnmf_for_five_paired_tag.mlx and jsnmf_for_three_mod.mlx, respectively.
The tutorial for inferring cell-type-specific region-gene associations is presented in example/Gene_Assoc_Tutorial.md.
If you have any questions about the source code, please feel free to contact me: chonghua_1983@126.com. The python implementation of JSNMF are also provided in "https://github.com/cuhklinlab/JSNMF_py".
Ma, Y., Sun, Z., Zeng, P., Zhang, W., & Lin, Z. (2022). JSNMF enables effective and accurate integrative analysis of single-cell multiomics data. Briefings in Bioinformatics, 23(3), bbac105.