Fig 1.
scMoMtF overall structure and task module diagram.
A scMoMtF uses the matched single-cell multi-omics data as the input to the model and the overall model framework is encoder-decoder-discriminator-classifier. B The tasks process of scMoMtF. C The research process for the interpretability of scMoMtF.
Fig 2.
Visualization and performance evaluation of dimension reduction task of scMoMtF compared with other comparison algorithms.
A-C Visualization of dimension reduction data generated by scMoMtF, Matilda, scMDC, and MultiVI on SNARE-seq, PBMC, and SHARE-seq datasets. D Visualization of dimension reduction data generated by scMoMtF, Matilda, scMDC and totalVI on the CITE-seq dataset. E-G Evaluate the clustering performance of dimension reduction data generated by scMoMtF, Matilda, scMDC, and MultiVI on SNARE-seq, PBMC, and SHARE-seq datasets using AMI, NMI, and ARI. H The clustering performance of dimension reduction data generated by scMoMtF, Matilda, scMDC and totalVI on CITE-seq dataset.
Fig 3.
Cell classification performance of scMoMtF.
A Comparison of classification accuracy between scMoMtF and other comparison algorithms under five-fold cross-validation. B The classification results of scMoMtF for each cell type in the PBMC dataset. C The classification results of scPred for each cell type in the PBMC dataset.
Fig 4.
scMoMtF single-cell multi-omics data simulation performance.
A-B scMoMtF visualizes the simulation effects of specified cell types on PBMC datasets. C scMoMtF visualizes the simulation effects of specified cell types on the CITE-seq dataset. D Pearson’s correlation between scMoMtF and other single-cell data simulation methods for highly variable genes in real and simulated data. Lower and upper hinges, first and third quartiles(Q1,Q3); whiskers, range of 1.5-times the interquartile; Centre line, median; Dot, outliers.
Fig 5.
scMoMtF corrects batch effect in the CITE-seq dataset.
A Visualization of selected data before removing batch effect. B Visualization of the results of batch correction by scMoMt using different batches. C The average classification accuracy across data batches of different batches.
Fig 6.
Interpretability analysis diagram of scMoMtF.
A Visualize the location of the cell subtype of CD8+ T cells in both the RNA and ATAC modalities. B The characteristics with high contribution in CD8+ T cells are normalized to a value between 0 and 1. C The expression degree of CD8B, CCL5 and GZMK in RNA modality and the expression degree of IL17C, LINC02446 and JAKMIP1 in ATAC modality. D Visualize the cell embedding of scMoMtF at different training periods using t-SNE.
Table 1.
Task training time (in seconds) of each method on different datasets.