Systems Biology (Methods in Molecular Biology)

has a benefit of reducing the number of hypothesis tests per-

formed for individual-gene-level differential expression analy-

sis) [20, 29]. Therefore, it is convenient to select (on a

per-species basis) the local intermodal minimum as the cutoff

for defining a gene that is expressed in at least one sample

group [30], as shown in the next step.

2. For each species, compute the minimum-expression-level cut-
   off using kernel density estimation. For the dog dataset, the R
   commands and expected output would be

maxexp_density_dog <- density(rsc_maxexp_dog$max_exp)
cutoff_dog <- optimize(approxfun(maxexp_density_dog$x, maxexp_density_dog$y),
interval=c(1,10))$minimum

cutoff_dog

[1] 2.96

(indicating an expression level cutoff of 2.96 for the dog

mRNA-seq dataset). In the above statement, the R function

"density" returns an R object representing the kernel density-

estimated distribution of the function argument, the R func-

tion "approxfun" performs linear interpolation, and the R

function "optimize" finds the point at which a given function

of one or more variables attains a minimum value over a speci-

fied region of the domain of the given function.

3. For each species, filter the data matrices to remove any genes
   that are not expressed based on the minimum-expression-level
   cutoff that was defined above. For the case of the dog dataset,
   the R commands and example output would be

0.0

0.2

0.4

0 5 10 15

max_exp

density

species

dog

human

Fig. 1Density distributions of the per-gene maximum (across sample groups) of

the within-sample-group-average log 2 expression level, for the dog and human

mRNA-seq data sets. It is clear that different expression level thresholds apply

for the two mRNA-seq data sets

Cross-Species RNA-Seq Analysis 297