Journal: Nature Communications

Article Title: Leveraging data-driven self-consistency for high-fidelity gene expression recovery

doi: 10.1038/s41467-022-34595-w

Figure Lengend Snippet: The histograms of the reference data, observed data (1% sampling efficiency), and imputed data by MAGIC, mcImpute, and SERM are shown in the first row of ( a ). Visualization of reference, observed, and imputed data by t-SNE and UMAP are shown in the second and third rows, respectively. t-SNE and UMAP results from SERM imputed data are much better in separating the classes, whereas MAGIC degrades the data due to imputation. The clustering accuracy and cluster quality indices for UMAP visualizations of imputed data from different methods are shown in ( b ). Data are presented as mean values +/− standard deviation (SD). Error bars represent the standard deviation of the indices for n = 1000 different initializations of k-means clustering. Source data are provided as a Source Data file.

Article Snippet: Other distributions can also be included in SERM (see Python/Matlab codes of SERM).

Techniques: Sampling, Standard Deviation

Journal: Nature Communications

Article Title: Leveraging data-driven self-consistency for high-fidelity gene expression recovery

doi: 10.1038/s41467-022-34595-w

Figure Lengend Snippet: UMAP results of the reference data, observed data, imputed data from MAGIC, mcImpute, and SERM for a cellular taxonomy, b mammalian brain, c mouse intestinal epithelium, and d 3D neural tissue data. Cellular taxonomy data was sampled at 10% efficiency, and the other three datasets were sampled at 0.1% efficiency. All the classes are better visualized in the SERM imputation. MAGIC and mcImpute distort the data in many cases, whereas SERM retains the consistency of the data intact in all cases. Source data are provided as a Source Data file.

Article Snippet: Other distributions can also be included in SERM (see Python/Matlab codes of SERM).

Techniques:

Journal: Nature Communications

Article Title: Leveraging data-driven self-consistency for high-fidelity gene expression recovery

doi: 10.1038/s41467-022-34595-w

Figure Lengend Snippet: PHATE results from the reference data (first column), observed data (second column), imputed data from MAGIC, mcImpute, and SERM (columns 3–5) for a zebrafish development data and b EB differentiation data. The observed data were created by sampling the reference data at 0.1% efficiency for both datasets. All the trajectories are better visualized in SERM imputed data. MAGIC and mcImpute distort the data in both cases, whereas SERM retains the consistency of the data intact in both cases. The colorbar for a denotes the hpf (hours post fertilization). The colorbar of b represents 1-(0–3 days), 2- (6–9 days), 3- (12–15 days), 4- (18–21 days) and 5- (24–27 days)). Pearson coefficient between the pseudotime estimated by monocle from the imputed data and the data labels for all the methods are shown for zebrafish development data (left), and EB differentiation data (right) in c . Source data are provided as a Source Data file.

Article Snippet: Other distributions can also be included in SERM (see Python/Matlab codes of SERM).

Techniques: Sampling