Data Upload

All files can be provided in csv, tsv and rds format. Vectors can be uploaded as txt as well.

Deconvolution Data

• Bulk RNA-seq data
  • Genes x Samples matrix
  • Rownames (gene names) are provided in the same format as in the sc RNA-seq data, for instance HGNC symbols
• Single cell RNA-seq data
  • Genes x Cells matrix
  • Counts are not log-transformed
  • Rownames (gene names) are provided in the same format as in the bulk RNA-seq data, for instance HGNC symbols
• Cell type annotations
  • Vector containing cell type annotations
  • Annotations are in the same order as the columns of the single cell matrix
• Batch ids
  • Vector containing batch ids, so sample or patient ids
  • Ids are in the same order as the columns of the single cell matrix
  • This is only necessary for Bisque, MuSiC and SCDC
• (Marker genes)
  • Vector containing gene names
  • This is only necessary for BSeq-sc

SimBu Simulation

If no ground truth data is available for your bulk dataset you can benchmark by simulating a pseudo-bulk sample with known cell type fraction using SimBu. To transfer your simulation to DeconvExplorer save it in rds format and upload it to retrieve the simulated bulk file and the corresponding ground truth.

simulation <- SimBu::simulate_bulk(...)
saveRDS(simulation, 'filepath.rds') # upload this file

Custom Signature and Ground Truth Reference

Please make sure the first column contains gene identifiers matching the bulk sample.

Deconvolution

Choose a deconvolution method to run a deconvolution with omnideconv. In total omnideconv offers 11 deconvolution algorithms. Each method builds a custom gene expression signature internally or, if applicable, you can choose to provide a custom one. The results can be visualized below and you can select multiple deconvolutions to be plotted at the same time. In addition you can inspect the results and available signatures in table form further below.

Signature Exploration

This module can visualize and compare the available signatures. Multiple metrics and plots are available. The prominently placed heatmap is interactive and allows you to select fragments of a signature for a more in depth analysis in the plot and the table below. To compare the gene sets of multiple signatures the UpSet Plot at the end of this module can be utilized.

Signature Refinement

If required this module offers four functions to further subselect genes and enhance signature quality. Before a refinement a signature should be loaded on the right side and needs to be saved by a new name afterwards. For benchmarking purposes it is also possible to rename cell types.

1. Remove Sequencing Artifacts Too many zeros in expression values of a gene across cell types can indicate a sequencing error. For this refinement you can choose the maximum percentage of zeros to be allowed for each gene. All other genes get discarded. It is recommended to use this function early in the refinement process as genes expressed in this manner get scored extraordinary high be the genes scores applied in the third function.

2. Remove genes expressed not specific enough This function works by dividing the expression range of each gene in three equal parts and assigning the expression values to one of the bins “high”, “medium” and “low”. With the parameter (default=1) you can select the maximum number of cell types a gene is allowed to be expressed in the “high” bin. For example, when running with the parameter set to 1 each gene is only allowed to be expressed in the highest bin in only 1 cell type. All other cell types need to fall into “medium” or “low” categories for this gene. All genes not fulfilling this requirement are removed from the signature.

3. Select best scored genes for each cell type For each cell type select a fixed amount of best scored genes. As scores Entropy and Gini Index can be applied. Both scores can be seen as heatmap annotation in the signature exploration and refinement module. The scores are used to quantify how cell type specific the expression of a gene is. The expression of a gene mainly in one cell type will result in low entropy values and a high Gini index.

Benchmark

To test deconvolution performance and compare methods or signatures the benchmarking module offers three different plots. For benchmarking a ground truth of the deconvoluted bulk sample is necessary and can be obtained from a SimBu simulation or by directly uploading a reference file in the Data Upload module. While only one reference file can be chosen multiple deconvolution results can be benchmarked at the same time.

This module offers a scatterplot visualizing the true vs estimated cell fractions. The result is better if it appears closer to the black line (identity line). In addition the correlation between the datasets and a Root Mean Square Error (RMSE) heatmap or boxplot can be displayed.

Further information

The code for omnideconv, SimBu and DeconvExplorer can be found on github and is linked above.

An interactive tour can be started in the menu bar at the top.