Normalization

Down-sampling

This strategy can be applied when the studied samples largely differ in their repertoire sizes.

By using the sampleRepSeqExp() function, users can choose the value to which all the samples are downsampled. If not specified, the lowest number of sequences across the dataset will be used.

This function returns a new RepSeqExperiment object with the downsized data.

RepSeqData_ds<- sampleRepSeqExp(x = RepSeqData, 
                                sample.size = 50000)
#> You set `rngseed` to FALSE. Make sure you've set & saved
#>   the random seed of your session for reproducibility.
#>  See `?set.seed`
#> Down-sampling to 50000 sequences...
#> Creating a RepSeqExperiment object...
#> Done.

Figure 1: Summary plots showing the effect of downsampling curves plotting the number of sequences and aaClones in each sample

Shannon-based normalization

This strategy adapted from Chaara et al., 2018 can be used to eliminate “uninformative” sequences resulting from experimental noise. It uses the Shannon entropy as a threshold and is applied at the ntClone level. This strategy is particularly efficient when applied on small samples as it corrects altered count distributions caused by a high-sequencing depth.

The function ShannonNorm() allows the application of this strategy without the need to specify any parameter and returns a new RepSeqExperiment object with the corrected data.

RepSeqData_sh <- ShannonNorm(x = RepSeqData)
#> Creating a RepSeqExperiment object...
#> Done.

Figure 2: A summary plot showing the number of ntClones pre-and post-Shannon normalizatin in each sample

Notes:

• Eliminated sequences in each sample are stored in the otherData slot.
• Chao, the richness estimator, isn’t recalculated for normalized datasets as as their original composition has been modified.