- Introduction -

Cell type fractions have been shown useful in many genomics analyses. Traditional methods for determining cell-type fractions like immunohistochemistry and flow cytometry remain costly compared to computational approaches using bulk RNA-seq data. Many computational methods, which are called cell type deconvolution, have been proposed to infer cell-type fractions from bulk transcriptomics data. However, these methods produce very different results under different settings.

We introduce EnsDeconv (Ensemble Deconvolution), which uses ensemble learning to to robustly estimate cellular fractions from bulk omics data.

---

- Get Started - Get Started

- Analysis - Analysis

Get Started

The deconv_ensemble app allow users to perform ensemble deconvolution as well as other deconvolution methods.

Analyze your data in the “Analysis” tab.

Instructions

Code can be found on github:

Please post issues on github, and feel free to contribute by forking and submitting development branches.

To run this app locally on your machine, download R or RStudio and run the following commands once to set up the environment:

To get detailed description of how to choose input, click:

Input Data

You may use this app by

Upload your own bulk data that is either i. RNA-seq data ii. Microarray data iii. DNA methylation data

Bulk Data and Reference Format

Can be a .CSV comma-separated-value file (you may export from Excel), .rds, and .txt file
File must have a header row.
First/Left-hand column(s) must be gene identifiers.

Count or Expression Data

Each row denotes a gene, each column denotes a sample.

Analysis: When raw counts are uploaded, the data is then analyzed by the app. The app allow user to choose data transformation and data scaling approaches.

The gene names between bulk data and reference data should be consistent or have intersection.

Meta Data for Reference Data

If you choose to upload reference data instead of using in-house datasets, you will also need to upload metadata for the reference data. The metadata should include sample names for the reference (column names of the reference) and deconvolution clusters, such as cell types.

Deconvolution important factors

For a detailed description of important factors in deconvolution, please refer to our original paper.

Output

You will find a boxplot, heatmap, and table displaying the estimated fraction from EnsDeconv.

Once you have chose all the paramters, click “Run” to start analysis.

You can also download estimated cell type fractions (including EnsDeconv and all other scenarios) by clicking “Download Results”:

Analysis

EnsDeconv output

Distribution of estimated fraction

Heatmap of estimated fraction

Estimated fraction

About

A user-friendly R Shiny app for ensemble cellular deconvolution to estimate cellular fractions from bulk omics, developed by Dr. Jiebiao Wang's group.

Creator: Manqi Cai.

Contributors: Liang You and Tianyuzhou (Jenny) Liang.

University of Pittsburgh

Instructions

Input Data

Bulk Data and Reference Format

Count or Expression Data

Meta Data for Reference Data

Deconvolution important factors

Output

Bulk Data

Custom Bulk Data

References

Ref Data

Meta data

Select cell type variable

Select sample ID variable

Choose tissue to inclue multiple references

Deconvolution Method

Marker Gene Approach

Num. of Marker Gene

Type of reference data

Type of scaling approach

Type of normalization approach

Parallel Computing

Num. of cores

Time Tolerance (minutes)