Transforming and calibrating LFQ data
Witold E. Wolski
Functional Genomics Center Zurich (FGCZ), ETH Zurich and University of ZurichSource:
vignettes/TransformingLFQData.Rmd
TransformingLFQData.RmdThe active response
LFQData holds the measurements and their configuration.
Its response() method gives the name of the intensity
column used by downstream operations. get_Transformer()
creates a transformer with its own copy of the data. Transformation
methods return the transformer for chaining, and $lfq
retrieves the result.
This example uses protein-level data from three samples. The two
standards, StdA and StdB, each have a constant
amount across samples. Their measured signals are doubled in
S2 and halved in S3 relative to
S1.
library(prolfqua)
samples <- c("S1", "S2", "S3")
quant <- data.frame(
protein_Id = rep(c("StdA", "StdB", "Target"), each = 3),
sampleName = rep(samples, 3),
raw.file = rep(paste0(samples, ".raw"), 3),
condition = rep(c("Control", "Control", "Treatment"), 3),
intensity = c(64, 128, 32, 256, 512, 128, 512, 1024, 512),
nr_children = 1
)
cfg <- AnalysisConfiguration$new()
cfg$file_name <- "raw.file"
cfg$hierarchy <- list(protein_Id = "protein_Id")
cfg$factors <- list(condition = "condition")
cfg$set_response("intensity")
lfq <- LFQData$new(quant, cfg, setup = TRUE)
logged <- lfq$get_Transformer()$log2()$lfq
logged$response()
#> [1] "log2_intensity"
head(logged$data_long()[c("protein_Id", "sampleName", lfq$response(), logged$response())])
#> # A tibble: 6 × 4
#> protein_Id sampleName intensity log2_intensity
#> <chr> <chr> <dbl> <dbl>
#> 1 StdA S1 64 6
#> 2 StdB S1 256 8
#> 3 Target S1 512 9
#> 4 StdA S2 128 7
#> 5 StdB S2 512 9
#> 6 Target S2 1024 10log2() expects positive measured intensities and carries
missing intensities through as NA. Convert zeros that
encode non-detection to NA before logging. The
is_transformed() flag records transformation status. Track
the scale separately when using an operation that requires a particular
scale.
| Method | Operation on the active response |
|---|---|
log2() |
Take the base-2 logarithm of each intensity. |
intensity_array(.func) |
Apply a function such as asinh to the response
vector. |
intensity_matrix(.func) |
Apply a matrix function to features in rows and samples in columns.
For example, intensity_matrix(vsn::justvsn) fits a
variance-stabilizing transformation to raw intensities. |
robscale() |
Centre each sample by its median and scale by its relative median absolute deviation. The default restores a common level based on the original medians. |
robscale_subset(lfqsubset) |
Estimate the medians and median absolute deviations from a subset of features and apply them to the full data. |
Calibrate against internal standards
Internal standards are reference proteins measured in the same
samples as the targets. Select proteins expected to have stable amounts
across samples. center_to_reference() subtracts the median
log2 intensity of the reference proteins from every feature in each
sample. The median of the reference proteins is then zero in each sample
with a usable reference. When all reference intensities for a sample are
missing, its centred intensities are missing.
reference_ids <- data.frame(protein_Id = c("StdA", "StdB"))
reference <- logged$get_subset(reference_ids)
calibrated_internal <- logged$get_Transformer()$center_to_reference(reference)$lfqIn prolfquapp, give the measured protein IDs in the DEA YAML under
processing_options$internal. For a protein-level analysis,
transform: none means a log2 transform without additional
global normalization. With internal configured, prolfquapp
computes a median from the finite log2 reference intensities per sample
and calls subtract_sample_offsets(). If a sample has no
finite reference intensity, it stops and lists the affected sample
IDs.
The VSN and robscale paths centre on their respective transformed scales. For nested analyses, prolfquapp uses the internal standards for the protein abundances shown in the report. The peptide model uses the transformed peptide data before that protein-level centring.
Apply offsets from external calibrants
Use subtract_sample_offsets() when the correction comes
from measurements outside the LFQ table. Supply one finite log2 offset
per sample, named by the configured sample ID. For example, suppose an
external calibrant has a reference signal of 128, and you have matched
one calibrant measurement to each sample. Divide each measured signal by
128 to obtain the fold change in response, then take its log2.
external <- data.frame(
sampleName = c("S3", "S1", "S2"),
measured_intensity = c(64, 128, 256)
)
offsets <- setNames(log2(external$measured_intensity / 128), external$sampleName)
offsets
#> S3 S1 S2
#> -1 0 1
calibrated_external <- logged$get_Transformer()$subtract_sample_offsets(offsets)$lfqThe order of the calibrant rows can differ from the order in the LFQ table. The method matches by sample ID and rejects missing, duplicate or unknown IDs. The caller is responsible for establishing that the external measurement tracks a multiplicative effect shared by the features in that sample.
If the log2 intensity is and the offset is , the calibrated intensity is . On the original scale this divides the measurement by . All features in a sample receive the same shift, which preserves within-sample log2 differences. Between-sample differences change by the difference between their offsets.
Compare the two calibrations
target_values <- function(lfq) {
rows <- lfq$data_long()
target <- rows[rows$protein_Id == "Target", ]
setNames(target[[lfq$response()]], target$sampleName)[samples]
}
knitr::kable(
data.frame(
Sample = samples,
`Original log2` = unname(target_values(logged)),
`Internal standards` = unname(target_values(calibrated_internal)),
`External calibrant` = unname(target_values(calibrated_external)),
check.names = FALSE
),
caption = "Target log2 intensity before and after calibration."
)| Sample | Original log2 | Internal standards | External calibrant |
|---|---|---|---|
| S1 | 9 | 2 | 9 |
| S2 | 10 | 2 | 9 |
| S3 | 9 | 3 | 10 |
Both calibrations recover a target that is equal in S1
and S2 and twice as abundant in S3. The
absolute log2 levels differ because internal centring puts the reference
median at zero, while the external offsets are anchored to the
calibrant’s reference signal of 128. The example treats them as two
estimates of the same sample effect, so it applies each to a separate
copy of the logged data.
Session info
sessionInfo()
#> R version 4.6.2 (2026-10-29)
#> Platform: x86_64-pc-linux-gnu
#> Running under: Ubuntu 26.04.1 LTS
#>
#> Matrix products: default
#> BLAS: /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3
#> LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.32.so; LAPACK version 3.12.0
#>
#> locale:
#> [1] LC_CTYPE=C.UTF-8 LC_NUMERIC=C LC_TIME=C.UTF-8
#> [4] LC_COLLATE=C.UTF-8 LC_MONETARY=C.UTF-8 LC_MESSAGES=C.UTF-8
#> [7] LC_PAPER=C.UTF-8 LC_NAME=C LC_ADDRESS=C
#> [10] LC_TELEPHONE=C LC_MEASUREMENT=C.UTF-8 LC_IDENTIFICATION=C
#>
#> time zone: UTC
#> tzcode source: system (glibc)
#>
#> attached base packages:
#> [1] stats graphics grDevices utils datasets methods base
#>
#> other attached packages:
#> [1] prolfqua_1.9.4 BiocStyle_2.40.0
#>
#> loaded via a namespace (and not attached):
#> [1] limma_3.68.5 gtable_0.3.6 jsonlite_2.0.0
#> [4] dplyr_1.2.1 compiler_4.6.2 BiocManager_1.30.27
#> [7] Rcpp_1.1.2 tidyselect_1.2.1 tidyr_1.3.2
#> [10] jquerylib_0.1.4 systemfonts_1.3.3 scales_1.4.0
#> [13] textshaping_1.0.5 statmod_1.5.2 yaml_2.3.12
#> [16] fastmap_1.2.0 ggplot2_4.0.4 R6_2.6.1
#> [19] generics_0.1.4 knitr_1.52 MASS_7.3-66
#> [22] htmlwidgets_1.6.4 ggrepel_0.9.8 tibble_3.3.1
#> [25] bookdown_0.48 desc_1.4.3 RColorBrewer_1.1-3
#> [28] bslib_0.12.0 pillar_1.11.1 rlang_1.3.0
#> [31] utf8_1.2.6 cachem_1.1.0 xfun_0.62
#> [34] S7_1.0.0 fs_2.1.0 sass_0.4.10
#> [37] otel_0.3.0 cli_3.6.6 withr_3.0.3
#> [40] pkgdown_2.2.1 magrittr_2.0.5 digest_0.6.39
#> [43] grid_4.6.2 lifecycle_1.0.5 vctrs_0.7.3
#> [46] evaluate_1.0.5 glue_1.8.1 farver_2.1.2
#> [49] ragg_1.5.2 purrr_1.2.2 rmarkdown_2.32
#> [52] tools_4.6.2 pkgconfig_2.0.3 htmltools_0.5.9