Generate synthetic count data sets

Share:

Description

Generate synthetic count data sets, following the simulation strategy detailed in Soneson and Delorenzi (2013).

Usage

1
2
3
4
5
6
7
8
generateSyntheticData(dataset, n.vars, samples.per.cond, n.diffexp,
  repl.id = 1, seqdepth = 1e+07, minfact = 0.7, maxfact = 1.4,
  relmeans = "auto", dispersions = "auto", fraction.upregulated = 1,
  between.group.diffdisp = FALSE, filter.threshold.total = 1,
  filter.threshold.mediancpm = 0, fraction.non.overdispersed = 0,
  random.outlier.high.prob = 0, random.outlier.low.prob = 0,
  single.outlier.high.prob = 0, single.outlier.low.prob = 0,
  effect.size = 1.5, output.file = NULL)

Arguments

dataset

A name or identifier for the data set/simulation settings.

n.vars

The initial number of genes in the simulated data set. Based on the filtering conditions (filter.threshold.total and filter.threshold.mediancpm), the number of genes in the final data set may be lower than this number.

samples.per.cond

The number of samples in each of the two conditions.

n.diffexp

The number of genes simulated to be differentially expressed between the two conditions.

repl.id

A replicate ID for the specific simulation instance. Useful for example when generating multiple count matrices with the same simulation settings.

seqdepth

The base sequencing depth (total number of mapped reads). This number is multiplied by a value drawn uniformly between minfact and maxfact for each sample to generate data with different actual sequencing depths.

minfact,maxfact

The minimum and maximum for the uniform distribution used to generate factors that are multiplied with seqdepth to generate individual sequencing depths for the simulated samples.

relmeans

A vector of mean values to use in the simulation of data from the Negative Binomial distribution, or "auto". Note that these values may be scaled in order to comply with the given sequencing depth. With the default value ("auto"), the mean values are sampled from values estimated from the Pickrell and Cheung data sets. If relmeans is a vector, the provided values will be used as mean values in the simulation for the samples in the first condition. The mean values for the samples in the second condition are generated by combining the relmeans and effect.size arguments.

dispersions

A vector or matrix of dispersions to use in the simulation of data from the Negative Binomial distribution, or "auto". With the default value ("auto"), the dispersion values are sampled from values estimated from the Pickrell and Cheung data sets. If both relmeans and dispersions are set to "auto", the means and dispersion values are sampled in pairs from the values in these data sets. If dispersions is a single vector, the provided dispersions will be used for simulating data from both conditions. If it is a matrix with two columns, the values in the first column are used for condition 1, and the values in the second column are used for condition 2.

fraction.upregulated

The fraction of the differentially expressed genes that is upregulated in condition 2 compared to condition 1.

between.group.diffdisp

Whether or not the dispersion should be allowed to be different between the conditions. Only applicable if dispersions is "auto".

filter.threshold.total

The filter threshold on the total count for a gene across all samples. All genes for which the total count across all samples is less than the threshold will be filtered out.

filter.threshold.mediancpm

The filter threshold on the median count per million (cpm) for a gene across all samples. All genes for which the median cpm across all samples is less than the threshold will be filtered out.

fraction.non.overdispersed

The fraction of the genes that should be simulated according to a Poisson distribution, without overdispersion. The non-overdispersed genes will be divided proportionally between the upregulated, downregulated and non-differentially expressed genes.

random.outlier.high.prob

The fraction of 'random' outliers with unusually high counts.

random.outlier.low.prob

The fraction of 'random' outliers with unusually low counts.

single.outlier.high.prob

The fraction of 'single' outliers with unusually high counts.

single.outlier.low.prob

The fraction of 'single' outliers with unusually low counts.

effect.size

The strength of the differential expression, i.e., the effect size, between the two conditions. If this is a single number, the effect sizes will be obtained by simulating numbers from an exponential distribution (with rate 1) and adding the results to the effect.size. For genes that are upregulated in the second condition, the mean in the first condition is multiplied by the effect size. For genes that are downregulated in the second condition, the mean in the first condition is divided by the effect size. It is also possible to provide a vector of effect sizes (one for each gene), which will be used as provided. In this case, the fraction.upregulated and n.diffexp arguments will be ignored and the values will be derived from the effect.size vector.

output.file

If not NULL, the path to the file where the data object should be saved. The extension should be .rds, if not it will be changed.

Details

In the comparison function, only results obtained for data sets with the same value of the dataset parameter will be compared. Hence, it is important to give the same value of this parameter e.g. to different replicates generated with the same simulation settings.

For more detailed information regarding the different types of outliers, see Soneson and Delorenzi (2013).

Mean and dispersion parameters (if relmeans and/or dispersions is set to "auto") are sampled from values estimated from the data sets by Pickrell et al (2010) and Cheung et al (2010). The data sets were downloaded from the ReCount web page (Frazee et al (2011)) and processed as detailed by Soneson and Delorenzi (2013).

To get the actual mean value for the Negative Binomial distribution used for the simulation of counts for a given sample, take the column truemeans.S1 (or truemeans.S2, if the sample is in condition S2) of the variable.annotations slot, divide by the sum of the same column and multiply with the base sequencing depth (provided in the info.parameters list) and the depth factor for the sample (given in the sample.annotations data frame). Thus, if you have a vector of mean values that you want to provide as the relmeans argument and make sure to use it 'as-is' in the simulation (for condition S1), make sure to set the seqdepth argument to the sum of the values in the relmeans vector, and to set minfact and maxfact equal to 1.

Value

A compData object. If output.file is not NULL, the object is saved in the given output.file (which should have an .rds extension).

Author(s)

Charlotte Soneson

References

Soneson C and Delorenzi M (2013): A comparison of methods for differential expression analysis of RNA-seq data. BMC Bioinformatics 14:91

Cheung VG, Nayak RR, Wang IX, Elwyn S, Cousins SM, Morley M and Spielman RS (2010): Polymorphic cis- and trans-regulation of human gene expression. PLoS Biology 8(9):e1000480

Frazee AC, Langmead B and Leek JT (2011): ReCount: a multi-experiment resource of analysis-ready RNA-seq gene count datasets. BMC Bioinformatics 12:449

Pickrell JK, Marioni JC, Pai AA, Degner JF, Engelhardt BE, Nkadori E, Veyrieras JB, Stephens M, Gilad Y and Pritchard JK (2010): Understanding mechanisms underlying human gene expression variation with RNA sequencing. Nature 464, 768-772

Robles JA, Qureshi SE, Stephen SJ, Wilson SR, Burden CJ and Taylor JM (2012): Efficient experimental design and analysis strategies for the detection of differential expression using RNA-sequencing. BMC Genomics 13:484

Examples

1
2
mydata.obj <- generateSyntheticData(dataset = "mydata", n.vars = 1000,
                                    samples.per.cond = 5, n.diffexp = 100)

Want to suggest features or report bugs for rdrr.io? Use the GitHub issue tracker.