InstaPrism

InstaPrism is an R package for cell type composition and gene expression deconvolution in bulk RNA-Seq data. Based on the same conceptual framework as BayesPrism, InstaPrism significantly increases calculation speed while providing nearly identical deconvolution results.

Installation

if (!require("BiocManager", quietly = TRUE))
    install.packages("BiocManager")

if (!require("Biobase", quietly = TRUE))
    BiocManager::install("Biobase")

if (!require("devtools", quietly = TRUE))
    install.packages("devtools")

devtools::install_github("humengying0907/InstaPrism")

Deconvolution with InstaPrism

Quick start with the following example

# bulk input with genes in rows and samples in columns; this example data is from ovarian cancer
bulk_expr = read.csv(system.file('extdata','example_bulk.csv',package = 'InstaPrism')) 

# load built-in reference; choose a reference that matches the tumor type of the bulk input
OV_ref = InstaPrism_reference('OV') 

# deconvolution with InstaPrism
deconv_res = InstaPrism(bulk_Expr = bulk_expr,refPhi_cs = OV_ref)

# The deconvoled cell type fraction is now accessible with
estimated_frac = t(deconv_res@Post.ini.ct@theta)
head(estimated_frac)

# InstaPrism also returns the deconvolved gene expression Z
Z = get_Z_array(deconv_res) # a sample by gene by cell-type array
head(Z[,1:10,'malignant'])

Deconvolution results comparison with BayesPrism

Using either scRNA-based reference (update = F) or updated reference (update = T), InstaPrism achieves nearly identical deconvolution results as BayesPrism.

Running time comparsion with BayesPrism

Below is a running time comparsion when running deconvolution on the tutorial data provided in BayesPrism.

Memory Comparison with BayesPrism

InstaPrism significantly reduced the memory usaged during deconvolution. Below is a memory usage comparsion when running deconvolution on the tutorial data provided in BayesPrism.

Built-in Reference for InstaPrism

We have provided precompiled reference tailored for a wide range of cancer types. The reference can be downloaded directly using the download link below, or loaded directly using the following command

BRCA_ref = InstaPrism_reference('BRCA')

reference name	tumor type	#cells used for reference construction	#cell types/cell states	umap	citation	download
BRCA_refPhi	breast cancer	100,064	8/76	UMAP	Wu et al. 2021	↓
CRC_refPhi	colorectal cancer	371,223	15/98	UMAP	Pelka et al. 2021	↓
GBM_refPhi	glioblastoma	338,564	10/57	cellxgeneLink, UMAP	Ruiz et al. 2022	↓
LUAD_refPhi	lung adenocarcinomas	118,293	13/77	UMAP	Xing et al. 2021	↓
OV_refPhi	ovarian cancer	929,690	9/40	cellxgeneLink, UMAP	Vazquez et al. 2022	↓
RCC_refPhi	clear cell renal cell carcinoma	270,855	11/106	cellxgeneLink, UMAP	Li et al. 2022	↓
SKCM_refPhi	skin cutaneous melanoma	4,645	8/23	UMAP	Tirosh et al. 2016	↓

Reference evaluation pipeline

To validate the performance of a reference, or to compare between different references, please refer to the Reference_evalu_pipeline we developed.

Tutorial

Check InstaPrism_tutorial for detailed implementation of InstaPrism and compare its performance with BayesPrism.

Reference

M. Hu and M. Chikina, “InstaPrism: an R package for fast implementation of BayesPrism.” bioRxiv, p. 2023.03.07.531579, Mar. 12, 2023. doi: https://doi.org/10.1101/2023.03.07.531579

humengying0907 / InstaPrism

readme