Resolution of the curse of dimensionality (RECODE) is a noise reduction method for single-cell sequencing data based on high-dimensional statistics.

Y. Imoto, T. Nakamura, et al. Resolution of the curse of dimensionality in single-cell RNA sequencing data analysis, 2022, Life Science Alliance.

Y. Imoto. Comprehensive Noise Reduction in Single-Cell Data with the RECODE Platform, 2024, bioRxiv.

The license gives permission for personal, academic, or educational use. Any commercial use is strictly prohibited. Please contact imoto.yusuke.4e@kyoto-u.ac.jp for licensing terms for any commercial use.

• Overview
• Python code
• R code
• R code (Python calling)
• Desktop Application
• License
• Contact

• Input is single-cell sequencing data (count matrix) $X \in \mathbb{Z}_{\geq 0}^{n\times d}$, where $n$ is the number of sample, $d$ is the number of features. For exmple, for scRNA-seq data, $n$ and $d$ correspond to the number of cells and genes, respectively.
• Compute the denoised data $X \in \mathbb{R}_{\geq 0}^{n\times d}$ with the same scale with $X$.
• Compute the applicability of RECODE, classified strongly applicable, weekly applicable, and inapplicable, denoting the level of accuracy of noise reduction.

To install RECODE package, use pip as follows:

$ pip install screcode

PyPi downloads (by PePy)

Tutorials and API reference

• Python3
• numpy
• scipy
• scikit-learn

Remark: The current version of the R code is not fast because of the lower speed of the PCA algorithm on R. So, we recommend using the Python code or R code with Python calling (below) for large-scale data.

You can install RECODE on R with:

devtools::install_github("yusuke-imoto-lab/RECODE/R")

For the single-cell sequeincing data X (rows: genes/epigenomes, columns: cells), we can apply RECODE as follows.

library(RECODE)

X.RECODE <- RECODE(X)

In the Seurat analysis, we can apply RECODE to SeuratObject and set it as default, as follows:

library(RECODE)
library(Matrix)

data <- as.matrix(seurat[["RNA"]]@counts)
data_RECODE <- RECODE(data)
seurat[["RECODE"]] <- CreateAssayObject(counts = Matrix(data_RECODE, sparse = TRUE))
DefaultAssay(seurat) <- "RECODE"

For a detailed analysis, please see below:

Tutorial (Run)

Tutorial (Run, QC, Clustering, Annotating etc.)

After installing remotes (install.packages("remotes")), you can install "recodeinstaller" with the following command:

remotes::install_github("yusuke-imoto-lab/recodeinstaller")

Then, the following command installs the Python version of RECODE.

library(recodeinstaller)

install_screcode()

Regarding the detail of installer, please see recodeinstaller.

For the single-cell sequeincing data X (rows: genes/epigenomes, columns: cells), we can apply RECODE as follows.

library(reticulate)

source("recodeloader/load_recodeenv.R")

plt <- reticulate::import(module="matplotlib.pyplot")
screcode <- reticulate::import(module="screcode.screcode")
X.RECODE<-screcode$RECODE(X)

Below is a more detailed analysis:

Tutorial (Python calling)

Installation and Tutorials

Windows (exe) and MAC OS (dmg) applications are avairable.

MIT © 2022 Yusuke Imoto

The license gives permission for personal, academic, or educational use. Any commercial use is strictly prohibited. Please contact imoto.yusuke.4e@kyoto-u.ac.jp for licensing terms for any commercial use.

• Yusuke Imoto
• Email: imoto.yusuke.4e@kyoto-u.ac.jp
• GitHub: yusuke-imoto-lab/RECODE