RSEQNORM

RSEQNORM implements MIXnorm and SMIXnorm for RNA-seq data normalization.

Click here for an online version and more normalization methods.

Introduction

MIXnorm and SMIXnorm are normalization methods designed for Formalin-Fixed Paraffin-Embedded (FFPE) RNA-sequencing (RNA-seq) data. MIXnorm relies on a two-component mixture model, which models non-expressed genes by zero-inflated Poisson distributions and models expressed genes by truncated normal distributions. SMIXnorm is a simplified version of MIXnorm, which uses a simplified mixture model and requires less computation. We recommend using SMIXnorm for FFPE RNA-seq data normalization for faster computation when the number of samples is larger than 25. Though designed specifically for FFPE RNA-seq data, MIXnorm and SMIXnorm are directly applicable to normalize fresh-frozen (FF) RNA-seq data as well. To obtain the maximum likelihood estimates, we developed a nested EM algorithm, in which closed-form updates are available in each iteration.

Dependencies

R version: > 3.5.3 (2019-3-11)

Platform: x86_64-apple-darwin15.6.0 (64-bit)

Running under: macOS Mojave 10.14.5

R Packages: truncnorm_1.0-8

Input files

Input file should be a J-by-I raw read count matrix in gene level with (J genes)*(I samples). We encourage to supply gene names as row names and sample names as column names. An example of the input data in R is shown below.

> dim(example.data)
[1] 18458    32
> example.data[1:6,1:6]
          NF9   TF9  NF10  TF10  NF11  TF11
A1BG        0     0     1     2     0     0
A1CF      595    67   292    52   259   112
A2M     45347 56829 15779 39418 33654 32715
A2ML1       2     0     1     3     0     0
A3GALT2     3     5     2     4     1     1
A4GALT    497   382   248   429   351   312

Example

Install

You can install RSEQNORM from github using the devtool.

install.packages("devtools")
library(devtools)
install_github("S-YIN/RSEQNORM")
library(RSEQNORM)

Example data

The ccRCC.RData is the clear cell renal cell carcinoma (ccRCC) data from Eikrem et al. (2016), which contains 18,458 protein coding genes and 32 FFPE RNA-seq data.

data(ccRCC)
head(ccRCC)

Usage

RSEQNORM is implemented in R. The scripts are under folder R. MIXnorm and SMIXnorm are the core functions that produce the normalized expression matrix ($MIX_normalized_log and $SMIX_normalized_log), proportion of expressed genes ($phi) and the probability of being expressed for each gene ($D).

Run SMIXnorm and MIXnorm

smix <- SMIXnorm(dat = ccRCC, max_iter = 20, tol = 0.01, appr = TRUE)
mix <- MIXnorm(dat = ccRCC, max_iter = 20, tol = 0.01, appr = TRUE)

See ?SMIXnorm and ?MIXnorm for additional documentations.

Obtain normalized expression

normalized.by.smix <- smix$SMIX_normalized_log
normalized.by.mix <- mix$MIX_normalized_log

The normalized expression is in natural logarithm scale. It is a matrix in the same dimension as the input matrix.

> dim(normalized.by.smix)
[1] 18458    32
> normalized.by.smix[1:6,1:6]
              NF9        TF9      NF10       TF10      NF11       TF11
A1BG    0.0000000  0.0000000 0.0000000  0.0000000 0.0000000  0.0000000
A1CF    0.8885373 -0.8772782 0.9724398 -1.1300976 0.5749542 -0.2046440
A2M     5.2204180  5.8510337 4.9587658  5.4816137 5.4381895  5.4635877
A2ML1   0.0000000  0.0000000 0.0000000  0.0000000 0.0000000  0.0000000
A3GALT2 0.0000000  0.0000000 0.0000000  0.0000000 0.0000000  0.0000000
A4GALT  0.7088968  0.8512491 0.8097201  0.9633957 0.8779038  0.8141714

Identify expressed genes

express.gene.smix <- rownames(ccRCC)[smix$D > 0.5]
express.gene.mix <- rownames(ccRCC)[mix$D > 0.5]

$D is the probability of being expressed for each gene. You may choose different cut-off value (the example uses 0.5 ) to identify expressed genes. $phi is the proportion of expressed genes among the data. In the example data, both MIXnorm and SMIXnorm identify 78.7% genes as being expressed.

> smix$D[1:6]
        A1BG         A1CF          A2M        A2ML1      A3GALT2       A4GALT 
6.562856e-47 1.000000e+00 1.000000e+00 1.410672e-48 1.538995e-27 1.000000e+00

> round(smix$phi,3)
[1] 0.787
> round(mix$phi,3)
[1] 0.787

Details

• The input data must be raw read count matrix of dimension J genes * I samples.
• The default maximum number of nested EM iteration (max_iter) is 20, recommend range (10, 50).
• The default convergency criteria (tol) is 0.01, recommend range (1e-5, 1).
• The default setting uses an approximate version of normalization (appr=TRUE). The exact normalization uses the posterior probabilities of genes being expressed or not to produce the normalized data. The approximate version uses a cut-off value of 0.5 on those probabilities to classify genes as expressed or not, then non-expressed genes will be normalized to exact 0.

Citation

Yin, S., Wang, X., Jia, G., and Xie, Y. (2020). MIXnorm: Normalizing Gene Expression Data from RNA Sequencing of Formalin-Fixed Paraffin-Embedded Samples. Bioinformatics. In Press. DOI: 10.1093/bioinformatics/btaa153.

Contact

Shen Yin (syin@smu.edu) Department of Statistical Science, Southern Methodist University.